Dissertations / Theses on the topic 'Classification rule'

Thesis (M.S.)--Worcester Polytechnic Institute.
Keywords: Itemset Pruning, Association Rules, Adaptive Minimal Support, Associative Classification, Classification. Includes bibliographical references (p.70-74).

The problem explored in this thesis is a supervised classification problem. Input data consists of operational and manufacturing data of a truck. The output denotes its operation, i.e. its basic utility and usage pattern, such as “Long distance” or “On and off-road”. In order to understand the distinction between the operation categories in practice, we look at interpretable classifiers. The examined classifiers are treeand rule-based classifiers, as they are the most interpretable. These include random forest, decision tree, and a classifier called inTrees, a method that summarizes a random forest using rules. In addition, a suggested method is examined. The suggested method works similarly to inTrees, but differs in the rule selection step. The question is whether this suggested method is better than inTrees in terms of interpretability, and how well both of them perform in comparison to a decision tree and a random forest. Another question regards the operation category of trucks, and whether they can be successfully distinguished using these methods.In order to compare the methods, their balanced accuracy, number of rules and other measures are recorded for the truck data set and additional data sets. Additional data sets are used to get a more exhaustive comparison between the methods.The suggested method does not outperform inTrees, and frequently uses three to four times as many rules to achieve the same accuracy on a given data set. Results indicate that the suggested method could perform more similarly to inTrees, given a different form of hyperparameter tuning. Additionally, it is shown that using interpretable classifiers rather than a random forest means we use less than one percent of the rules, at the cost of a loss of 10 percentage points in balanced accuracy.
Problemet som utforskas i detta examensarbete är ett problem inom övervakat lärande där indata består av driftdata samt tillverkningsspecifikationer för en lastbil, och utdata är dess användningsområde, såsom “Långdistans” eller “Stadsdistribution”. Målet är att få insikt i vad distinktionen mellan lastbilars användningsområden är i praktiken. För att utreda detta används regeloch trädbaserade klassificerare. Dessa används eftersom de är de mest tolkningsbara klassificerarna. De klassificerare som ingår är random forest, beslutsträd och en klassificerare kallad inTrees, som extraherar regler från en random forest. Utöver detta föreslås en ny metod som bygger på inTrees, men som skiljer sig i hur den väljer regler.Frågeställningen är om den föreslagna metoden ger resultat av högre tolkningsbarhet än inTrees, och hur väl bägge presterar i jämförelse med ett beslutsträd och en random forest. En annan del av frågeställningen gäller vad för slutsatser som kan dras kring användningsområde av lastbilar.För att jämföra prestandan av dessa metoder undersöktes både prediktionsgraden och tolkningsbarheten. Detta gjordes för lastbilsdatat men även andra publika dataset. Andra dataset användes för att få en mer omfattande jämförelse.Den föreslagna metoden är mindre tolkningsbar än inTrees då den ofta kräver tre till fyra gånger så många regler för att uppnå samma precision för ett dataset. Vissa resultat indikerar att den föreslagna metoden kunnat prestera mer likt inTrees om en annan hyperparameter-optimisering hade använts. Ytterligare resultat visade att vi, genom att använda tolkningsbara klassificerare istället för random forest, förlorade 10 procentenheter i balanserad precision men använde mindre än en procent av reglerna.

The purpose of the proposed thesis is to:1. Define expert systems and discuss various implementation techniques for the components of expert systems. This includes discussion on knowledge representation, inference methods, methods for dealing with uncertainty, and methods of explanation. Specifically, the focus will be on the implementation of rule-based expert systems;2. Apply selected expert system techniques to a case study. The case study will be a rule-based expert system in Prolog to recognize and identify musical chords from tonal harmony. The system will have a general knowledge base containing fundamental rules about chord construction. It will also contain some knowledge that will allow it to deduce non-trivial chords. Furthermore, it will contain procedures to deal with uncertainty and explanation;3. Explain general concepts about music theory and tonal chord classification to put the case study in context; and4. Discuss the limitations of expert systems based on the results of the case study and the current literature.
Department of Computer Science

Classification using association rules is a research field in data mining that primarily uses association rule discovery techniques in classification benchmarks. It has been confirmed by many research studies in the literature that classification using association tends to generate more predictive classification systems than traditional classification data mining techniques like probabilistic, statistical and decision tree. In this thesis, we introduce a novel data mining algorithm based on classification using association called “Looking at the Class” (LC), which can be used in for mining a range of classification data sets. Unlike known algorithms in classification using the association approach such as Classification based on Association rule (CBA) system and Classification based on Predictive Association (CPAR) system, which merge disjoint items in the rule learning step without anticipating the class label similarity, the proposed algorithm merges only items with identical class labels. This saves too many unnecessary items combining during the rule learning step, and consequently results in large saving in computational time and memory. Furthermore, the LC algorithm uses a novel prediction procedure that employs multiple rules to make the prediction decision instead of a single rule. The proposed algorithm has been evaluated thoroughly on real world security data sets collected using an automated tool developed at Huddersfield University. The security application which we have considered in this thesis is about categorizing websites based on their features to legitimate or fake which is a typical binary classification problem. Also, experimental results on a number of UCI data sets have been conducted and the measures used for evaluation is the classification accuracy, memory usage, and others. The results show that LC algorithm outperformed traditional classification algorithms such as C4.5, PART and Naïve Bayes as well as known classification based association algorithms like CBA with respect to classification accuracy, memory usage, and execution time on most data sets we consider.

This thesis introduces a unified framework for design of rule based systems for classification tasks, which consists of the operations of rule generation, rule simplification and rule representation. This thesis also stresses the importance of combination of different rule learning algorithms through ensemble learning approaches. For the three operations mentioned above, novel approaches are developed and validated by comparing with existing ones for advancing the performance of using this framework. In particular, for rule generation, Information Entropy Based Rule Generation is developed and validated through comparing with Prism. For rule simplification, Jmid-pruning is developed and validated through comparing with J-pruning and Jmax-pruning. For rule representation, rule based network is developed and validated through comparing with decision tree and linear list. The results show that the novel approaches complement well the existing ones in terms of accuracy, efficiency and interpretability. On the other hand, this thesis introduces ensemble learning approaches that involve collaborations in training or testing stage through combination of learning algorithms or models. In particular, the novel framework Collaborative and Competitive Random Decision Rules is created and validated through comparing with Random Prisms. This thesis also introduces the other novel framework Collaborative Rule Generation which involves collaborations in training stage through combination of multiple learning algorithms. This framework is validated through comparing with each individual algorithm. In addition, this thesis shows that the above two frameworks can be combined as a hybrid ensemble learning framework toward advancing overall performance of classification. This hybrid framework is validated through comparing with Random Forests. Finally, this thesis summarises the research contributions in terms of theoretical significance, practical importance, methodological impact and philosophical aspects. In particular, theoretical significance includes creation of the framework for design of rule based systems and development of novel approaches relating to rule based classification. Practical importance shows the usefulness in knowledge discovery and predictive modelling and the independency in application domains and platforms. Methodological impact shows the advances in generation, simplification and representation of rules. Philosophical aspects include the novel understanding of data mining and machine learning in the context of human research and learning, and the inspiration from information theory, system theory and control theory toward methodological innovations. On the basis of the completed work, this thesis provides suggestions regarding further directions toward advancing this research area.

There is growing evidence that integrating classification and association rule mining can produce more efficient and accurate classifiers than traditional techniques. This thesis introduces a new MapReduce based association rule miner for extracting strong rules from large datasets. This miner is used later to develop a new large scale classifier. Also new MapReduce simulator was developed to evaluate the scalability of proposed algorithms on MapReduce clusters. The developed associative rule miner inherits the MapReduce scalability to huge datasets and to thousands of processing nodes. For finding frequent itemsets, it uses hybrid approach between miners that uses counting methods on horizontal datasets, and miners that use set intersections on datasets of vertical formats. The new miner generates same rules that usually generated using apriori-like algorithms because it uses the same confidence and support thresholds definitions. In the last few years, a number of associative classification algorithms have been proposed, i.e. CPAR, CMAR, MCAR, MMAC and others. This thesis also introduces a new MapReduce classifier that based MapReduce associative rule mining. This algorithm employs different approaches in rule discovery, rule ranking, rule pruning, rule prediction and rule evaluation methods. The new classifier works on multi-class datasets and is able to produce multi-label predications with probabilities for each predicted label. To evaluate the classifier 20 different datasets from the UCI data collection were used. Results show that the proposed approach is an accurate and effective classification technique, highly competitive and scalable if compared with other traditional and associative classification approaches. Also a MapReduce simulator was developed to measure the scalability of MapReduce based applications easily and quickly, and to captures the behaviour of algorithms on cluster environments. This also allows optimizing the configurations of MapReduce clusters to get better execution times and hardware utilization.

The rapidly growing volume and complexity of modern databases makes the need for technologies to describe and summarise the information they contain increasingly important. Data mining is a process of extracting implicit, previously unknown and potentially useful patterns and relationships from data, and is widely used in industry and business applications. Rules characterise relationships among patterns in databases, and rule mining is one of the central tasks in data mining. There are fundamentally two categories of rules, namely association rules and classification rules. Traditionally, association rules are connected with transaction databases for market basket problems and classification rules are associated with relational databases for predictions. In this thesis, we will mainly focus on the use of association rules for predictions. An optimal rule set is a rule set that satisfies given optimality criteria. In this thesis we study two types of optimal rule sets, the informative association rule set and the optimal class association rule set, where the informative association rule set is used for market basket predictions and the class association rule set is used for the classification. A robust classification rule set is a rule set that is capable of providing more correct predictions than a traditional classification rule set on incomplete test data. Mining transaction databases for association rules usually generates a large number of rules, most of which are unnecessary when used for subsequent prediction. We define a rule set for a given transaction database that is significantly smaller than an association rule set but makes the same predictions as the complete association rule set. We call this rule set the informative rule set. The informative rule set is not constrained to particular target items; and it is smaller than the non-redundant association rule set. We characterise the relationships between the informative rule set and the non-redundant association rule set. We present an algorithm to directly generate the informative rule set without generating all frequent itemsets first, and that accesses databases less often than other direct methods. We show experimentally that the informative rule set is much smaller than both the association rule set and the non-redundant association rule set for a given database, and that it can be generated more efficiently. In addition, we discuss a new unsupervised discretization method to deal with numerical attributes in general association rule mining without target specification. Based on the analysis of the strengths and weaknesses of two commonly used unsupervised numerical attribute discretization methods, we present an adaptive numerical attribute merging algorithm that is shown better than both methods in general association rule mining. Relational databases are usually denser than transaction databases, so mining on them for class association rules, which is a set of association rules whose consequences are classes, may be difficult due to the combinatorial explosion. Based on the analysis of the prediction mechanism, we define an optimal class association rule set to be a subset of the complete class association rule set containing all potentially predictive rules. Using this rule set instead of the complete class association rule set we can avoid redundant computation that would otherwise be required for mining predictive association rules and hence improve the efficiency of the mining process significantly. We present an efficient algorithm for mining optimal class association rule sets using upward closure properties to prune weak rules before they are actually generated. We show theoretically the efficiency of the proposed algorithm will be greater than Apriori on dense databases, and confirm experimentally that it generates an optimal class association rule set, which is very much smaller than a complete class association rule set, in significantly less time than generating the complete class association rule set by Apriori. Traditional classification rule sets perform badly on test data that are not as complete as the training data. We study the problem of discovering more robust rule sets, i.e. we say a rule is more robust than another rule set if it is able to make more accurate predictions on test data with missing attribute values. We reveal a hierarchy of k-optimal rule sets where a k-optimal rule set with a large k is more robust, and they are more robust than a traditional classification rule set. We introduce two methods to find k-optimal rule sets, i.e. an optimal association rule mining approach and a heuristic approximate approach. We show experimentally that a k-optimal rule set generated from the optimal association rule mining approach performs better than that from the heuristic approximate approach and both rule sets perform significantly better than a typical classification rule set (C4.5Rules) on incomplete test data. Finally, we summarise the work discussed in this thesis, and suggest some future research directions.
Thesis (PhD Doctorate)
Doctor of Philosophy (PhD)
School of Computing and Information Technology
Science, Environment, Engineering and Technology
Full Text

Recent studies in data mining revealed that Associative Classification (AC) data mining approach builds competitive classification classifiers with reference to accuracy when compared to classic classification approaches including decision tree and rule based. Nevertheless, AC algorithms suffer from a number of known defects as the generation of large number of rules which makes it hard for end-user to maintain and understand its outcome and the possible over-fitting issue caused by the confidence-based rule evaluation used by AC. This thesis attempts to deal with above problems by presenting five new pruning methods, prediction method and employs them in an AC algorithm that significantly reduces the number of generated rules without having large impact on the prediction rate of the classifiers. Particularly, the new pruning methods that discard redundant and insignificant rules during building the classifier are employed. These pruning procedures remove any rule that either has no training case coverage or covers a training case without the requirement of class similarity between the rule class and that of the training case. This enables large coverage for each rule and reduces overfitting as well as construct accurate and moderated size classifiers. Beside, a novel class assignment method based on multiple rules is proposed which employs group of rule to make the prediction decision. The integration of both the pruning and prediction procedures has been used to enhanced a known AC algorithm called Multiple-class Classification based on Association Rules (MCAR) and resulted in competent model in regard to accuracy and classifier size called " Multiple-class Classification based on Association Rules 2(MCAR2)". Experimental results against different datasets from the UCI data repository showed that the predictive power of the resulting classifiers in MCAR2 slightly increase and the resulting classifier size gets reduced comparing with other AC algorithms such as Multiple-class Classification based on Association Rules (MCAR).

Data mining is the process of discovering useful patterns from datasets using intelligent techniques to help users make certain decisions. A typical data mining task is classification, which involves predicting a target variable known as the class in previously unseen data based on models learnt from an input dataset. Covering is a well-known classification approach that derives models with If-Then rules. Covering methods, such as PRISM, have a competitive predictive performance to other classical classification techniques such as greedy, decision tree and associative classification. Therefore, Covering models are appropriate decision-making tools and users favour them carrying out decisions. Despite the use of Covering approach in data processing for different classification applications, it is also acknowledged that this approach suffers from the noticeable drawback of inducing massive numbers of rules making the resulting model large and unmanageable by users. This issue is attributed to the way Covering techniques induce the rules as they keep adding items to the rule’s body, despite the limited data coverage (number of training instances that the rule classifies), until the rule becomes with zero error. This excessive learning overfits the training dataset and also limits the applicability of Covering models in decision making, because managers normally prefer a summarised set of knowledge that they are able to control and comprehend rather a high maintenance models. In practice, there should be a trade-off between the number of rules offered by a classification model and its predictive performance. Another issue associated with the Covering models is the overlapping of training data among the rules, which happens when a rule’s classified data are discarded during the rule discovery phase. Unfortunately, the impact of a rule’s removed data on other potential rules is not considered by this approach. However, When removing training data linked with a rule, both frequency and rank of other rules’ items which have appeared in the removed data are updated. The impacted rules should maintain their true rank and frequency in a dynamic manner during the rule discovery phase rather just keeping the initial computed frequency from the original input dataset. In response to the aforementioned issues, a new dynamic learning technique based on Covering and rule induction, that we call Enhanced Dynamic Rule Induction (eDRI), is developed. eDRI has been implemented in Java and it has been embedded in WEKA machine learning tool. The developed algorithm incrementally discovers the rules using primarily frequency and rule strength thresholds. These thresholds in practice limit the search space for both items as well as potential rules by discarding any with insufficient data representation as early as possible resulting in an efficient training phase. More importantly, eDRI substantially cuts down the number of training examples scans by continuously updating potential rules’ frequency and strength parameters in a dynamic manner whenever a rule gets inserted into the classifier. In particular, and for each derived rule, eDRI adjusts on the fly the remaining potential rules’ items frequencies as well as ranks specifically for those that appeared within the deleted training instances of the derived rule. This gives a more realistic model with minimal rules redundancy, and makes the process of rule induction efficient and dynamic and not static. Moreover, the proposed technique minimises the classifier’s number of rules at preliminary stages by stopping learning when any rule does not meet the rule’s strength threshold therefore minimising overfitting and ensuring a manageable classifier. Lastly, eDRI prediction procedure not only priorities using the best ranked rule for class forecasting of test data but also restricts the use of the default class rule thus reduces the number of misclassifications. The aforementioned improvements guarantee classification models with smaller size that do not overfit the training dataset, while maintaining their predictive performance. The eDRI derived models particularly benefit greatly users taking key business decisions since they can provide a rich knowledge base to support their decision making. This is because these models’ predictive accuracies are high, easy to understand, and controllable as well as robust, i.e. flexible to be amended without drastic change. eDRI applicability has been evaluated on the hard problem of phishing detection. Phishing normally involves creating a fake well-designed website that has identical similarity to an existing business trustful website aiming to trick users and illegally obtain their credentials such as login information in order to access their financial assets. The experimental results against large phishing datasets revealed that eDRI is highly useful as an anti-phishing tool since it derived manageable size models when compared with other traditional techniques without hindering the classification performance. Further evaluation results using other several classification datasets from different domains obtained from University of California Data Repository have corroborated eDRI’s competitive performance with respect to accuracy, number of knowledge representation, training time and items space reduction. This makes the proposed technique not only efficient in inducing rules but also effective.

COORDENAÇÃO DE APERFEIÇOAMENTO DO PESSOAL DE ENSINO SUPERIOR
Esta dissertação investiga a utilização de sistemas Neuro- Fuzzy Hierárquicos BSP (Binary Space Partitioning) para classificação de padrões e para extração de regras fuzzy em bases de dados. O objetivo do trabalho foi criar modelos específicos para classificação de registros a partir do modelo Neuro-Fuzzy Hierárquico BSP que é capaz de gerar sua própria estrutura automaticamente e extrair regras fuzzy, lingüisticamente interpretáveis, que explicam a estrutura dos dados. O princípio da tarefa de classificação de padrões é descobrir relacionamentos entre os dados com a intenção de prever a classe de um padrão desconhecido. O trabalho consistiu fundamentalmente de quatro partes: um estudo sobre os principais métodos de classificação de padrões; análise do sistema Neuro-Fuzzy Hierárquico BSP (NFHB) original na tarefa de classificação; definição e implementação de dois sistemas NFHB específicos para classificação de padrões; e o estudo de casos. No estudo sobre os métodos de classificação foi feito um levantamento bibliográfico da área, resultando em um "survey" onde foram apresentadas as principais técnicas utilizadas para esta tarefa. Entre as principais técnicas destacaram-se: os métodos estatísticos, algoritmos genéticos, árvores de decisão fuzzy, redes neurais, e os sistemas neuro-fuzzy. Na análise do sistema NFHB na classificação de dados levou- se em consideração as peculiaridades do modelo, que possui: aprendizado da estrutura, particionamento recursivo do espaço de entrada, aceita maior número de entradas que os outros sistemas neuro-fuzzy, além de regras fuzzy recursivas. O sistema NFHB, entretanto, não é um modelo exatamente desenvolvido para classificação de padrões. O modelo NFHB original possui apenas uma saída e para utilizá- lo como um classificador é necessário criar um critério de faixa de valores (janelas) para representar as classes. Assim sendo, decidiu-se criar novos modelos que suprissem essa deficiência. Foram definidos dois novos sistemas NFHB para classificação de padrões: NFHB-Invertido e NFHB-Class. O primeiro utiliza a arquitetura do modelo NFHB original no aprendizado e em seguida a inversão da mesma para a validação dos resultados. A inversão do sistema consistiu de um meio de adaptar o novo sistema à tarefa específica de classificação, pois passou-se a ter o número de saídas do sistema igual ao número de classes ao invés do critério de faixa de valores utilizado no modelo NFHB original. Já o sistema NFHB-Class utilizou, tanto para a fase de aprendizado, quanto para a fase de validação, o modelo NFHB original invertido. Ambos os sistemas criados possuem o número de saídas igual ao número de classes dos padrões, o que representou um grande diferencial em relação ao modelo NFHB original. Além do objetivo de classificação de padrões, o sistema NFHB-Class foi capaz de extrair conhecimento em forma de regras fuzzy interpretáveis. Essas regras são expressas da seguinte maneira: SE x é A e y é B então padrão pertence à classe Z. Realizou-se um amplo estudo de casos, abrangendo diversas bases de dados Benchmark para a tarefa de classificação, tais como: Iris Dataset, Wine Data, Pima Indians Diabetes Database, Bupa Liver Disorders e Heart Disease, e foram feitas comparações com diversos modelos e algoritmos de classificação de padrões. Os resultados encontrados com os modelos NFHB-Invertido e NFHB-Class mostraram-se, na maioria dos casos, superiores ou iguais aos melhores resultados encontrados pelos outros modelos e algoritmos aos quais foram comparados.O desempenho dos modelos NFHB-Invertido e NFHB-Class em relação ao tempo de processamento também se mostrou muito bom. Para todas as bases de dados descritas no estudo de casos (capítulo 8), os modelos convergiram para uma ótima solução de classificação, além da extração das regras fuzzy, em
This dissertation investigates the use of Neuro-Fuzzy Hierarchical BSP (Binary Space Partitioning) systems for pattern classification and extraction of fuzzy rules in databases. The objective of this work was to create specific models for the classification of registers based on the Neuro-Fuzzy BSP model that is able to create its structure automatically and to extract linguistic rules that explain the data structure. The task of pattern classification is to find relationships between data with the intention of forecasting the class of an unknown pattern. The work consisted of four parts: study about the main methods of the pattern classification; evaluation of the original Neuro-Fuzzy Hierarchical BSP system (NFHB) in pattern classification; definition and implementation of two NFHB systems dedicated to pattern classification; and case studies. The study about classification methods resulted in a survey on the area, where the main techniques used for pattern classification are described. The main techniques are: statistic methods, genetic algorithms, decision trees, neural networks, and neuro-fuzzy systems. The evaluation of the NFHB system in pattern classification took in to consideration the particularities of the model which has: ability to create its own structure; recursive space partitioning; ability to deal with more inputs than other neuro-fuzzy system; and recursive fuzzy rules. The original NFHB system, however, is unsuited for pattern classification. The original NFHB model has only one output and its use in classification problems makes it necessary to create a criterion of band value (windows) in order to represent the classes. Therefore, it was decided to create new models that could overcome this deficiency. Two new NFHB systems were developed for pattern classification: NFHB-Invertido and NFHB-Class. The first one creates its structure using the same learning algorithm of the original NFHB system. After the structure has been created, it is inverted (see chapter 5) for the generalization process. The inversion of the structure provides the system with the number of outputs equal to the number of classes in the database. The second system, the NFHB-Class uses an inverted version of the original basic NFHB cell in both phases, learning and validation. Both systems proposed have the number of outputs equal to the number of the pattern classes, what means a great differential in relation to the original NFHB model. Besides the pattern classification objective, the NFHB- Class system was able to extract knowledge in form of interpretable fuzzy rules. These rules are expressed by this way: If x is A and y is B then the pattern belongs to Z class. The two models developed have been tested in many case studies, including Benchmark databases for classification task, such as: Iris Dataset, Wine Data, Pima Indians Diabetes Database, Bupa Liver Disorders and Heart Disease, where comparison has been made with several traditional models and algorithms of pattern classification. The results found with NFHB-Invertido and NFHB-Class models, in all cases, showed to be superior or equal to the best results found by the others models and algorithms for pattern classification. The performance of the NFHB- Invertido and NFHB-Class models in terms of time-processing were also very good. For all databases described in the case studies (chapter 8), the models converged to an optimal classification solution, besides the fuzzy rules extraction, in a time-processing inferior to a minute.
Esta disertación investiga el uso de sistemas Neuro- Fuzzy Herárquicos BSP (Binary Space Partitioning) en problemas de clasificación de padrones y de extracción de reglas fuzzy en bases de datos. El objetivo de este trabajo fue crear modelos específicos para clasificación de registros a partir del modelo Neuro-Fuzzy Jerárquico BSP que es capaz de generar automáticamente su propia extructura y extraer reglas fuzzy, lingüisticamente interpretables, que explican la extructura de los datos. El principio de la clasificación de padrones es descubrir relaciones entre los datos con la intención de prever la clase de un padrón desconocido. El trabajo está constituido por cuatro partes: un estudio sobre los principales métodos de clasificación de padrones; análisis del sistema Neuro-Fuzzy Jerárquico BSP (NFHB) original en la clasificación; definición e implementación de dos sistemas NFHB específicos para clasificación de padrones; y el estudio de casos. En el estudio de los métodos de clasificación se realizó un levatamiento bibliográfico, creando un "survey" donde se presentan las principales técnicas utilizadas. Entre las principales técnicas se destacan: los métodos estadísticos, algoritmos genéticos, árboles de decisión fuzzy, redes neurales, y los sistemas neuro-fuzzy. En el análisis del sistema NFHB para clasificación de datos se tuvieron en cuenta las peculiaridades del modelo, que posee : aprendizaje de la extructura, particionamiento recursivo del espacio de entrada, acepta mayor número de entradas que los otros sistemas neuro-fuzzy, además de reglas fuzzy recursivas. El sistema NFHB, sin embargo, no es un modelo exactamente desarrollado para clasificación de padrones. El modelo NFHB original posee apenas una salida y para utilizarlo conmo un clasificador fue necesario crear un criterio de intervalos de valores (ventanas) para representar las clases. Así, se decidió crear nuevos modelos que supriman esta deficiencia. Se definieron dos nuevos sistemas NFHB para clasificación de padrones: NFHB- Invertido y NFHB-Clas. El primero utiliza la arquitectura del modelo NFHB original en el aprendizaje y en seguida la inversión de la arquitectura para la validación de los resultados. La inversión del sistema es un medio para adaptar el nuevo sistema, específicamente a la clasificación, ya que el sistema pasó a tener número de salidas igual al número de clases, al contrario del criterio de intervalo de valores utilizado en el modelo NFHB original. En el sistema NFHB-Clas se utilizó, tanto para la fase de aprendizajeo, cuanto para la fase de validación, el modelo NFHB original invertido. Ambos sistemas poseen el número de salidas igual al número de clases de los padrones, lo que representa una gran diferencia en relación al modelo NFHB original. Además del objetivo de clasificación de padrones, el sistema NFHB-Clas fue capaz de extraer conocimento en forma de reglas fuzzy interpretables. Esas reglas se expresan de la siguiente manera: Si x es A e y es B entonces el padrón pertenece a la clase Z. Se realizó un amplio estudio de casos, utilizando diversas bases de datos Benchmark para la clasificación, tales como: Iris Dataset, Wine Data, Pima Indians Diabetes Database, Bupa Liver Disorders y Heart Disease. Los resultados se compararon con diversos modelos y algoritmos de clasificación de padrones. Los resultados encontrados con los modelos NFHB-Invertido y NFHB-Clas se mostraron, en la mayoría de los casos, superiores o iguales a los mejores resultados encontrados por los otros modelos y algoritmos con los cuales fueron comparados. El desempeño de los modelos NFHB-Invertido y NFHB-Clas en relación al tiempo de procesamiento tambiém se mostró muy bien. Para todas las bases de datos descritas en el estudio de casos (capítulo 8), los modelos convergieron para una solución óptima, además de la extracción de las reglas fuzzy, con tiemp

Thesis (MSc (Geography and Environmental Studies))--University of Stellenbosch, 2010.
ENGLISH ABSTRACT: Supervised classifiers are the most popular approach for image classification due to their high accuracies, ease of use and strong theoretical grounding. Their primary disadvantage is the high level of user input required during the creation of the data needed to train the classifier. One alternative to supervised classification is an expert-system rule-based approach where expert knowledge is used to create a set of rules which can be applied to multiple images. This research compared supervised and expert-system rule-based approaches for forest mapping. For this purpose two SPOT 5 images were acquired and atmospherically corrected. Field visits, aerial photography, high resolution imagery and expert forestry knowledge were used for the compilation of the training data and the development of a rule-set. Both approaches were evaluated in an object-orientated environment. It was found that the accuracy of the resulting maps was equivalent, with both techniques returning an overall classification accuracy of 90%. This suggests that cost-effectiveness is the decisive factor for determining which method is superior. Although the development of the rule-set was time-consuming and challenging, it did not require any training data. In contrast, the supervised approach required a large number of training areas for each image classified, which was time-consuming and costly. Significantly more training areas will be required when the technique is applied to large areas, especially when multiple images are used. It was concluded that the rule-set is more cost-effective when applied at regional scale, but it is not viable for mapping small areas.
AFRIKAANSE OPSOMMING: Gerigte klassifiseerders is die gewildste benadering tot beeldklassifikasie as gevolg van hulle hoë graad van akkuraatheid, maklike aanwending en kragtige teoretiese fundering. Die primere nadeel van gerigte klassifikasie is die hoë vlak van gebruikersinsette wat benodig word tydens die skepping van opleidingsdata. 'n Alternatief vir gerigte klassifikasie is 'n deskundige stelsel waarin ‘n reëlgebaseerde benadering gevolg word om deskundige kennis aan te wend vir die opstel van 'n stel reëls wat op meervoudige beelde toegepas kan word. Hierdie navorsing het gerigte en deskundige stelsel benaderings toegepas vir bosboukartering om die twee benaderings met mekaar te vergelyk. Vir dié doel is twee SPOT 5 beelde verkry en atmosferies gekorrigeer. Veldbesoeke, lugfotografie, hoë-resolusie beelde en deskundige bosboukennis is aangewend om opleidingsdata saam te stel en die stel reëls te ontwikkel. Beide benaderings is in 'n objekgeoriënteerde omgewing beoordeel. Die akkuraatheidsvlakke van die resulterende kaarte was ewe hoog vir beide tegnieke met 'n algehele klassifikasie-akkuraatheid van 90%. Dit wil dus voorkom asof koste-effektiwiteit eerder as akkuraatheid die deurslaggewende faktor is om te bepaal watter metode die beste is. Alhoewel die ontwikkeling van die stel reëls tydrowend en uitdagend was, het dit geen opleidingsdata vereis nie. In teenstelling hiermee is 'n groot aantal opleidingsgebiede geskep vir elke beeld wat met gerigte klassifikasie verwerk is – 'n tydrowende en duur opsie. Dit is duidelik dat meer opleidingsgebiede benodig sal word wanneer die tegniek op groot gebiede toegepas word, veral omdat meervoudige beelde gebruik sal word. Gevolglik sal die stel reëls meer kosteeffektief wees wanneer dit op streekskaal toegepas word. ‘n Deskundige stelsel benadering is egter nie lewensvatbaar vir die kartering van klein gebiede nie.

Twitter is a microblogging application used for sending and retrieving instant on-line messages of not more than 140 characters. There has been a surge in Twitter activities since its launch in 2006 as well as steady increase in event detection research on Twitter data (tweets) in recent years. With 284 million monthly active users Twitter has continued to grow both in size and activity. The network is rapidly changing the way global audience source for information and influence the process of journalism [Newman, 2009]. Twitter is now perceived as an information network in addition to being a social network. This explains why traditional news media follow activities on Twitter to enhance their news reports and news updates. Knowing the significance of the network as an information dissemination platform, news media subscribe to Twitter accounts where they post their news headlines and include the link to their on-line news where the full story may be found. Twitter users in some cases, post breaking news on the network before such news are published by traditional news media. This can be ascribed to Twitter subscribers' nearness to location of events. The use of Twitter as a network for information dissemination as well as for opinion expression by different entities is now common. This has also brought with it the issue of computational challenges of extracting newsworthy contents from Twitter noisy data. Considering the enormous volume of data Twitter generates, users append the hashtag (#) symbol as prefix to keywords in tweets. Hashtag labels describe the content of tweets. The use of hashtags also makes it easy to search for and read tweets of interest. The volume of Twitter streaming data makes it imperative to derive Topic Detection and Tracking methods to extract newsworthy topics from tweets. Since hashtags describe and enhance the readability of tweets, this research is developed to show how the appropriate use of hashtags keywords in tweets can demonstrate temporal evolvements of related topic in real-life and consequently enhance Topic Detection and Tracking on Twitter network. We chose to apply our method on Twitter network because of the restricted number of characters per message and for being a network that allows sharing data publicly. More importantly, our choice was based on the fact that hashtags are an inherent component of Twitter. To this end, the aim of this research is to develop, implement and validate a new approach that extracts newsworthy topics from tweets' hashtags of real-life topics over a specified period using Association Rule Mining. We termed our novel methodology Transaction-based Rule Change Mining (TRCM). TRCM is a system built on top of the Apriori method of Association Rule Mining to extract patterns of Association Rules changes in tweets hashtag keywords at different periods of time and to map the extracted keywords to related real-life topic or scenario. To the best of our knowledge, the adoption of dynamics of Association Rules of hashtag co-occurrences has not been explored as a Topic Detection and Tracking method on Twitter. The application of Apriori to hashtags present in tweets at two consecutive period t and t + 1 produces two association rulesets, which represents rules evolvement in the context of this research. A change in rules is discovered by matching every rule in ruleset at time t with those in ruleset at time t + 1. The changes are grouped under four identified rules namely 'New' rules, 'Unexpected Consequent' and 'Unexpected Conditional' rules, 'Emerging' rules and 'Dead' rules. The four rules represent different levels of topic real-life evolvements. For example, the emerging rule represents very important occurrence such as breaking news, while unexpected rules represents unexpected twist of event in an on-going topic. The new rule represents dissimilarity in rules in rulesets at time t and t+1. Finally, the dead rule represents topic that is no longer present on the Twitter network. TRCM revealed the dynamics of Association Rules present in tweets and demonstrates the linkage between the different types of rule dynamics to targeted real-life topics/events. In this research, we conducted experimental studies on tweets from different domains such as sports and politics to test the performance effectiveness of our method. We validated our method, TRCM with carefully chosen ground truth. The outcome of our research experiments include: Identification of 4 rule dynamics in tweets' hashtags namely: New rules, Emerging rules, Unexpected rules and 'Dead' rules using Association Rule Mining. These rules signify how news and events evolved in real-life scenario. Identification of rule evolvements on Twitter network using Rule Trend Analysis and Rule Trace. Detection and tracking of topic evolvements on Twitter using Transaction-based Rule Change Mining TRCM. Identification of how the peculiar features of each TRCM rules affect their performance effectiveness on real datasets.

This thesis seeks to establish if the use of negation in Inductive Rule Learning (IRL) for text classification is effective. Text classification is a widely research topic in the domain of data mining. There have been many techniques directed at text classification; one of them is IRL, widely chosen because of its simplicity, comprehensibility and interpretability by humans. IRL is a process whereby rules in the form of $antecedent -> conclusion$ are learnt to build a classifier. Thus, the learnt classifier comprises a set of rules, which are used to perform classification. To learn a rule, words from pre-labelled documents, known as features, are selected to be used as conjunctions in the rule antecedent. These rules typically do not include any negated features in their antecedent; although in some cases, as demonstrated in this thesis, the inclusion of negation is required and beneficial for the text classification task. With respect to the use of negation in IRL, two issues need to be addressed: (i) the identification of the features to be negated and (ii) the improvisation of rule refinement strategies to generate rules both with and without negation. To address the first issue, feature space division is proposed, whereby the feature space containing features to be used for rule refinement is divided into three sub-spaces to facilitate the identification of the features which can be advantageously negated. To address the second issue, eight rule refinement strategies are proposed, which are able to generate both rules with and without negation. Typically, single keywords which are deemed significant to differentiate between classes are selected to be used in the text representation in the text classification task. Phrases have also been proposed because they are considered to be semantically richer than single keywords. Therefore, with respect to the work conducted in this thesis, three different types of phrases ($n$-gram phrases, keyphrases and fuzzy phrases) are extracted to be used as the text representation in addition to the use of single keywords. To establish the effectiveness of the use of negation in IRL, the eight proposed rule refinement strategies are compared with one another, using keywords and the three different types of phrases as the text representation, to determine whether the best strategy is one which generates rules with negation or without negation. Two types of classification tasks are conducted; binary classification and multi-class classification. The best strategy in the proposed IRL mechanism is compared to five existing text classification techniques with respect to binary classification: (i) the Sequential Minimal Optimization (SMO) algorithm, (ii) Naive Bayes (NB), (iii) JRip, (iv) OlexGreedy and (v) OlexGA from the Waikato Environment for Knowledge Analysis (WEKA) machine learning workbench. In the multi-class classification task, the proposed IRL mechanism is compared to the Total From Partial Classification (TFPC) algorithm. The datasets used in the experiments include three text datasets: 20 Newsgroups, Reuters-21578 and Small Animal Veterinary Surveillance Network (SAVSNET) datasets and five UCI Machine Learning Repository tabular datasets. The results obtained from the experiments showed that the strategies which generated rules with negation were more effective when the keyword representation was used and less prominent when the phrase representations were used. Strategies which generated rules with negation also performed better with respect to binary classification compared to multi-class classification. In comparison with the other machine learning techniques selected, the proposed IRL mechanism was shown to generally outperform all the compared techniques and was competitive with SMO.

Data mining is about the search for relationships and global patterns in large databases that are increasing in size. Data mining is beneficial for anyone who has a huge amount of data, for example, customer and business data, transaction, marketing, financial, manufacturing and web data etc. The results of data mining are also referred to as knowledge in the form of rules, regularities and constraints. Rule mining is one of the popular data mining methods since rules provide concise statements of potentially important information that is easily understood by end users and also actionable patterns. At present rule mining has received a good deal of attention and enthusiasm from data mining researchers since rule mining is capable of solving many data mining problems such as classification, association, customer profiling, summarization, segmentation and many others. This thesis makes several contributions by proposing rule mining methods using genetic algorithms and neural networks. The thesis first proposes rule mining methods using a genetic algorithm. These methods are based on an integrated framework but capable of mining three major classes of rules. Moreover, the rule mining processes in these methods are controlled by tuning of two data mining measures such as support and confidence. The thesis shows how to build data mining predictive models using the resultant rules of the proposed methods. Another key contribution of the thesis is the proposal of rule mining methods using supervised neural networks. The thesis mathematically analyses the Widrow-Hoff learning algorithm of a single-layered neural network, which results in a foundation for rule mining algorithms using single-layered neural networks. Three rule mining algorithms using single-layered neural networks are proposed for the three major classes of rules on the basis of the proposed theorems. The thesis also looks at the problem of rule mining where user guidance is absent. The thesis proposes a guided rule mining system to overcome this problem. The thesis extends this work further by comparing the performance of the algorithm used in the proposed guided rule mining system with Apriori data mining algorithm. Finally, the thesis studies the Kohonen self-organization map as an unsupervised neural network for rule mining algorithms. Two approaches are adopted based on the way of self-organization maps applied in rule mining models. In the first approach, self-organization map is used for clustering, which provides class information to the rule mining process. In the second approach, automated rule mining takes the place of trained neurons as it grows in a hierarchical structure.

Building an accurate and reliable model for prediction for different application domains, is one of the most significant challenges in knowledge discovery and data mining. This thesis focuses on building and enhancing a generic predictive model for estimating a future value by extracting association rules (knowledge) from a quantitative database. This model is applied to several data sets obtained from different benchmark problems, and the results are evaluated through extensive experimental tests. The thesis presents an incremental development process for the prediction model with three stages. Firstly, a Knowledge Discovery (KD) model is proposed by integrating Fuzzy C-Means (FCM) with Apriori approach to extract Fuzzy Association Rules (FARs) from a database for building a Knowledge Base (KB) to predict a future value. The KD model has been tested with two road-traffic data sets. Secondly, the initial model has been further developed by including a diversification method in order to improve a reliable FARs to find out the best and representative rules. The resulting Diverse Fuzzy Rule Base (DFRB) maintains high quality and diverse FARs offering a more reliable and generic model. The model uses FCM to transform quantitative data into fuzzy ones, while a Multiple Support Apriori (MSapriori) algorithm is adapted to extract the FARs from fuzzy data. The correlation values for these FARs are calculated, and an efficient orientation for filtering FARs is performed as a post-processing method. The FARs diversity is maintained through the clustering of FARs, based on the concept of the sharing function technique used in multi-objectives optimization. The best and the most diverse FARs are obtained as the DFRB to utilise within the Fuzzy Inference System (FIS) for prediction. The third stage of development proposes a hybrid prediction model called Fuzzy Associative Classification Rule Mining (FACRM) model. This model integrates the ii improved Gustafson-Kessel (G-K) algorithm, the proposed Fuzzy Associative Classification Rules (FACR) algorithm and the proposed diversification method. The improved G-K algorithm transforms quantitative data into fuzzy data, while the FACR generate significant rules (Fuzzy Classification Association Rules (FCARs)) by employing the improved multiple support threshold, associative classification and vertical scanning format approaches. These FCARs are then filtered by calculating the correlation value and the distance between them. The advantage of the proposed FACRM model is to build a generalized prediction model, able to deal with different application domains. The validation of the FACRM model is conducted using different benchmark data sets from the University of California, Irvine (UCI) of machine learning and KEEL (Knowledge Extraction based on Evolutionary Learning) repositories, and the results of the proposed FACRM are also compared with other existing prediction models. The experimental results show that the error rate and generalization performance of the proposed model is better in the majority of data sets with respect to the commonly used models. A new method for feature selection entitled Weighting Feature Selection (WFS) is also proposed. The WFS method aims to improve the performance of FACRM model. The prediction performance is improved by minimizing the prediction error and reducing the number of generated rules. The prediction results of FACRM by employing WFS have been compared with that of FACRM and Stepwise Regression (SR) models for different data sets. The performance analysis and comparative study show that the proposed prediction model provides an effective approach that can be used within a decision support system.

Building an accurate and reliable model for prediction for different application domains, is one of the most significant challenges in knowledge discovery and data mining. This thesis focuses on building and enhancing a generic predictive model for estimating a future value by extracting association rules (knowledge) from a quantitative database. This model is applied to several data sets obtained from different benchmark problems, and the results are evaluated through extensive experimental tests. The thesis presents an incremental development process for the prediction model with three stages. Firstly, a Knowledge Discovery (KD) model is proposed by integrating Fuzzy C-Means (FCM) with Apriori approach to extract Fuzzy Association Rules (FARs) from a database for building a Knowledge Base (KB) to predict a future value. The KD model has been tested with two road-traffic data sets. Secondly, the initial model has been further developed by including a diversification method in order to improve a reliable FARs to find out the best and representative rules. The resulting Diverse Fuzzy Rule Base (DFRB) maintains high quality and diverse FARs offering a more reliable and generic model. The model uses FCM to transform quantitative data into fuzzy ones, while a Multiple Support Apriori (MSapriori) algorithm is adapted to extract the FARs from fuzzy data. The correlation values for these FARs are calculated, and an efficient orientation for filtering FARs is performed as a post-processing method. The FARs diversity is maintained through the clustering of FARs, based on the concept of the sharing function technique used in multi-objectives optimization. The best and the most diverse FARs are obtained as the DFRB to utilise within the Fuzzy Inference System (FIS) for prediction. The third stage of development proposes a hybrid prediction model called Fuzzy Associative Classification Rule Mining (FACRM) model. This model integrates the ii improved Gustafson-Kessel (G-K) algorithm, the proposed Fuzzy Associative Classification Rules (FACR) algorithm and the proposed diversification method. The improved G-K algorithm transforms quantitative data into fuzzy data, while the FACR generate significant rules (Fuzzy Classification Association Rules (FCARs)) by employing the improved multiple support threshold, associative classification and vertical scanning format approaches. These FCARs are then filtered by calculating the correlation value and the distance between them. The advantage of the proposed FACRM model is to build a generalized prediction model, able to deal with different application domains. The validation of the FACRM model is conducted using different benchmark data sets from the University of California, Irvine (UCI) of machine learning and KEEL (Knowledge Extraction based on Evolutionary Learning) repositories, and the results of the proposed FACRM are also compared with other existing prediction models. The experimental results show that the error rate and generalization performance of the proposed model is better in the majority of data sets with respect to the commonly used models. A new method for feature selection entitled Weighting Feature Selection (WFS) is also proposed. The WFS method aims to improve the performance of FACRM model. The prediction performance is improved by minimizing the prediction error and reducing the number of generated rules. The prediction results of FACRM by employing WFS have been compared with that of FACRM and Stepwise Regression (SR) models for different data sets. The performance analysis and comparative study show that the proposed prediction model provides an effective approach that can be used within a decision support system.
Applied Science University (ASU) of Jordan

Thesis (MSc)--Stellenbosch University, 2005.
ENGLISH ABSTRACT: Remote sensing and image processing tools provide speedy and up-to-date information on land resources. Although remote sensing is the most effective means of land cover and land use mapping, it is not without limitations. The accuracy of image analysis depends on a number of factors, of which the image classifier used is probably the most significant. It is noted that there is no perfect classifier, but some robust classifiers achieve higher accuracy results than others. For certain land cover/uses, discrimination based only on spectral properties is extremely difficult and often produces poor results. The use of ancillary data can improve the classification process. Some classifiers incorporate ancillary data before or after the classification process, which limits the full utilization of the information contained in the ancillary data. Expert classification, on the other hand, makes better use of ancillary data by incorporating data directly into the classification process. In this study an expert classification model was developed based on spatial operations designed to identify a specific land cover/use, by integrating both spectral and available ancillary data. Ancillary data were derived either from the spectral channels or from other spatial data sources such as DEM (Digital Elevation Model) and topographical maps. The model was developed in ERDAS Imagine image-processing software, using the expert engineer as a final integrator of the different constituent spatial operations. An attempt was made to identify the Level I land cover classes in the South African National Land Cover classification scheme hierarchy. Rules were determined on the basis of expert knowledge or statistical calculations of mean and variance on training samples. Although rules could be determined by using statistical applications, such as the classification analysis regression tree (CART), the absence of adequate and accurate training data for all land cover classes and the fact that all land cover classes do not require the same predictor variables makes this option less desirable. The result of the accuracy assessment showed that the overall classification accuracy was 84.3% and kappa statistics 0.829. Although this level of accuracy might be suitable for most applications, the model is flexible enough to be improved further.
AFRIKAANSE OPSOMMING: Afstandswaameming-en beeldverwerkingstegnieke kan akkurate informasie oorbodemhulpbronne weergee. Alhoewel afstandswaameming die mees effektiewe manier van grondbedekking en grondgebruikkartering is, is dit nie sonder beperkinge nie. Die akkuraatheid van beeldverwerking is afhanklik van verskeie faktore, waarvan die beeld klassifiseerder wat gebruik word, waarskynlik die belangrikste faktor is. Dit is welbekend dat daar geen perfekte klassifiseerder is nie, alhoewel sekere kragtige klassifiseerders hoër akkuraatheid as ander behaal. Vir sekere grondbedekking en -gebruike is uitkenning gebaseer op spektrale eienskappe uiters moeilik en dikwels word swak resultate behaal. Die gebruik van aanvullende data, kan die klassifikasieproses verbeter. Sommige klassifiseerders inkorporeer aanvullende data voor of na die klassifikasieproses, wat die volle aanwending van die informasie in die aanvullende data beperk. Deskundige klassifikasie, aan die ander kant, maak beter gebruik van aanvullende data deurdat dit data direk in die klassifikasieproses inkorporeer. Tydens hierdie studie is 'n deskundige klassifikasiemodel ontwikkel gebaseer op ruimtelike verwerkings, wat ontwerp is om spesifieke grondbedekking en -gebruike te identifiseer. Laasgenoemde is behaal deur beide spektrale en beskikbare aanvullende data te integreer. Aanvullende data is afgelei van, óf spektrale eienskappe, óf ander ruimtelike bronne soos 'n DEM (Digitale Elevasie Model) en topografiese kaarte. Die model is ontwikkel in ERDAS Imagine beeldverwerking sagteware, waar die 'expert engineer' as finale integreerder van die verskillende samestellende ruimtelike verwerkings gebruik is. 'n Poging is aangewend om die Klas I grondbedekkingklasse, in die Suid-Afrikaanse Nasionale Grondbedekking klassifikasiesisteem te identifiseer. Reëls is vasgestel aan die hand van deskundige begrippe of eenvoudige statistiese berekeninge van die gemiddelde en variansie van opleidingsdata. Alhoewel reëls met behulp van statistiese toepassings, soos die 'classification analysis regression tree (CART)' vasgestel kon word, maak die afwesigheid van genoegsame en akkurate opleidingsdata vir al die grondbedekkingsklasse hierdie opsie minder aantreklik. Bykomend tot laasgenoemde, vereis alle grondbedekkingsklasse nie dieselfde voorspellingsveranderlikes nie. Die resultaat van hierdie akkuraatheidsskatting toon dat die algehele klassifikasie-akkuraatheid 84.3% was en die kappa statistieke 0.829. Alhoewel hierdie vlak van akkuraatheid vir die meeste toepassings geskik is, is die model aanpasbaar genoeg om verder te verbeter.

A rule-based classification model was developed to derive land-cover information from a large set of hyper-temporal, multi-spectral satellite imagery encompassing the state of Arizona. The model uses Advanced Very High Resolution Radiometer (AVHRR) imagery and the 30-minute digital elevation model (DEM) from the EROS Data Center (EDC) Conterminous U.S. AVHRR Biweekly Composites. Sixty one images from 1990, 1991 and 1992 were analyzed using the Brown & Lowe (1973) Natural Vegetative Communities of Arizona map to identify temporal patterns of Normalized Difference Vegetation Index (NDVI) and thermal measurements for 13 land-cover classes. Fifteen characteristic layers were created to represent the spectral, thermal and temporal properties of the data set. These layers were inputs for the rule-based classification model. The model was run on three years of data, creating three single year land-cover maps. The modeling effort showed that NDVI, thermal and DEM characteristics are useful for discerning land-cover classes. The single year land-cover maps showed that the rule-based model could not detect land-cover change between years. The single year maps were combined to create a summary land-cover map. This map differs from the Brown and Lowe map in the shape, proportional size and spatial distribution of land-cover polygons. The rule-based model can discern more land-cover classes than spectral cluster classification. Ground observations and an aerial video was used to assess map accuracy. The same proportion of agreement was observed between the ground observations, the Brown and Lowe map, and the summary land-cover map. Agreement was higher between video and the summary map than between video and the Brown and Lowe map. With further refinements to the input data set, classification model rules and field accuracy assessment, higher levels of agreement can be expected. Overall results show that rule-based classification of hyper-temporal, multi-spectral satellite imagery is a desirable method for mapping global land-cover.

Associative Classification (AC) in data mining is a rule based approach that uses association rule techniques to construct accurate classification systems (classifiers). The majority of existing AC algorithms extract one class per rule and ignore other class labels even when they have large data representation. Thus, extending current AC algorithms to find and extract multi-label rules is promising research direction since new hidden knowledge is revealed for decision makers. Furthermore, the exponential growth of rules in AC has been investigated in this thesis aiming to minimise the number of candidate rules, and therefore reducing the classifier size so end-user can easily exploit and maintain it. Moreover, an investigation to both rule ranking and test data classification steps have been conducted in order to improve the performance of AC algorithms in regards to predictive accuracy. Overall, this thesis investigates different problems related to AC not limited to the ones listed above, and the results are new AC algorithms that devise single and multi-label rules from different applications data sets, together with comprehensive experimental results. To be exact, the first algorithm proposed named Multi-class Associative Classifier (MAC): This algorithm derives classifiers where each rule is connected with a single class from a training data set. MAC enhanced the rule discovery, rule ranking, rule filtering and classification of test data in AC. The second algorithm proposed is called Multi-label Classifier based Associative Classification (MCAC) that adds on MAC a novel rule discovery method which discovers multi-label rules from single label data without learning from parts of the training data set. These rules denote vital information ignored by most current AC algorithms which benefit both the end-user and the classifier's predictive accuracy. Lastly, the vital problem related to web threats called 'website phishing detection' was deeply investigated where a technical solution based on AC has been introduced in Chapter 6. Particularly, we were able to detect new type of knowledge and enhance the detection rate with respect to error rate using our proposed algorithms and against a large collected phishing data set. Thorough experimental tests utilising large numbers of University of California Irvine (UCI) data sets and a variety of real application data collections related to website classification and trainer timetabling problems reveal that MAC and MCAC generates better quality classifiers if compared with other AC and rule based algorithms with respect to various evaluation measures, i.e. error rate, Label-Weight, Any-Label, number of rules, etc. This is mainly due to the different improvements related to rule discovery, rule filtering, rule sorting, classification step, and more importantly the new type of knowledge associated with the proposed algorithms. Most chapters in this thesis have been disseminated or under review in journals and refereed conference proceedings.

Dans de nombreux problèmes de classification, les données sont intrinsèquement incertaines. Les données d’apprentissage disponibles peuvent être imprécises, incomplètes, ou même peu fiables. En outre, des connaissances spécialisées partielles qui caractérisent le problème de classification peuvent également être disponibles. Ces différents types d’incertitude posent de grands défis pour la conception de classifieurs. La théorie des fonctions de croyance fournit un cadre rigoureux et élégant pour la représentation et la combinaison d’une grande variété d’informations incertaines. Dans cette thèse, nous utilisons cette théorie pour résoudre les problèmes de classification des données incertaines sur la base de deux approches courantes, à savoir, la méthode des k plus proches voisins (kNN) et la méthode à base de règles.Pour la méthode kNN, une préoccupation est que les données d’apprentissage imprécises dans les régions où les classes de chevauchent peuvent affecter ses performances de manière importante. Une méthode d’édition a été développée dans le cadre de la théorie des fonctions de croyance pour modéliser l’information imprécise apportée par les échantillons dans les régions qui se chevauchent. Une autre considération est que, parfois, seul un ensemble de données d’apprentissage incomplet est disponible, auquel cas les performances de la méthode kNN se dégradent considérablement. Motivé par ce problème, nous avons développé une méthode de fusion efficace pour combiner un ensemble de classifieurs kNN couplés utilisant des métriques couplées apprises localement. Pour la méthode à base de règles, afin d’améliorer sa performance dans les applications complexes, nous étendons la méthode traditionnelle dans le cadre des fonctions de croyance. Nous développons un système de classification fondé sur des règles de croyance pour traiter des informations incertains dans les problèmes de classification complexes. En outre, dans certaines applications, en plus de données d’apprentissage, des connaissances expertes peuvent également être disponibles. Nous avons donc développé un système de classification hybride fondé sur des règles de croyance permettant d’utiliser ces deux types d’information pour la classification
In many classification problems, data are inherently uncertain. The available training data might be imprecise, incomplete, even unreliable. Besides, partial expert knowledge characterizing the classification problem may also be available. These different types of uncertainty bring great challenges to classifier design. The theory of belief functions provides a well-founded and elegant framework to represent and combine a large variety of uncertain information. In this thesis, we use this theory to address the uncertain data classification problems based on two popular approaches, i.e., the k-nearest neighbor rule (kNN) andrule-based classification systems. For the kNN rule, one concern is that the imprecise training data in class over lapping regions may greatly affect its performance. An evidential editing version of the kNNrule was developed based on the theory of belief functions in order to well model the imprecise information for those samples in over lapping regions. Another consideration is that, sometimes, only an incomplete training data set is available, in which case the ideal behaviors of the kNN rule degrade dramatically. Motivated by this problem, we designedan evidential fusion scheme for combining a group of pairwise kNN classifiers developed based on locally learned pairwise distance metrics.For rule-based classification systems, in order to improving their performance in complex applications, we extended the traditional fuzzy rule-based classification system in the framework of belief functions and develop a belief rule-based classification system to address uncertain information in complex classification problems. Further, considering that in some applications, apart from training data collected by sensors, partial expert knowledge can also be available, a hybrid belief rule-based classification system was developed to make use of these two types of information jointly for classification

This paper investigates how an organization attempts to repair trust after organizational-level integrity violations by examining the influence of organizational rules on trust repair. We reconstruct the prominent corruption case of Siemens AG, which has faced the greatest bribery scandal in the history of German business. Our findings suggest that tightening organizational rules is an appropriate signal of trustworthiness for external stakeholders to demonstrate that the organization seriously intends to prevent integrity violations in the future. However, such rule adjustments were the source of dissatisfaction among employees since the new rules were difficult to implement in practice. We argue that these different impacts of organizational rules result from their inherent paradoxical nature. To address this problem, we suggest managing an effective interplay between formal and informal rules.

Text classification is to classify documents into predefined categories by learned classifiers. Classic text classifiers cannot unambiguously describe decision boundary between relevant and irrelevant documents because of uncertainties caused by feature selection and knowledge learning. This research proposes a three-way decision model for dealing with uncertain decision boundary based on rough sets and centroid solution to improve classification performance. It partitions training samples into three regions by two main boundary vectors, and resolves the boundary region by two derived boundary vectors to generate decision rules for making 'two-way' decisions.

Wo und wie werden komplexe Regelsätze im Gehirn repräsentiert? Drei empirische Studien dieser Doktorarbeit untersuchen dies experimentell. Eine weitere methodische Studie liefert Beiträge zur Weiterentwicklung der genutzten empirischen Methode. Die empirischen Studien nutzen multivariate Musteranalyse (MVPA) funktioneller Magnetresonanzdaten (fMRT) gesunder Probanden. Die Fragestellungen der methodischen Studie wurden durch die empirischen Arbeiten inspiriert. Wirkung und Anwendungsbreite der entwickelten Methode gehen jedoch über die Anwendung in den empirischen Studien dieser Arbeit hinaus. Die empirischen Studien bearbeiten Fragen wie: Wo werden Hinweisreize und Regeln repräsentiert, und sind deren Repräsentationen voneinander unabhängig? Wo werden Regeln repräsentiert, die aus mehreren Einzelregeln bestehen, und sind Repräsentationen der zusammengesetzten Regeln Kombinationen der Repräsentationen der Einzelregeln? Wo sind Regeln verschiedener Hierarchieebenen repräsentiert, und gibt es einen hierarchieabhängigen Gradienten im ventrolateralen präfrontalen Kortex (VLPFK)? Wo wird die Reihenfolge der Regelausführung repräsentiert? Alle empirischen Studien verwenden informationsbasiertes funktionales Mapping ("Searchlight"-Ansatz), zur hirnweiten und räumlich Lokalisierung von Repräsentationen verschiedener Elemente komplexer Regelsätze. Kernergebnisse der Arbeit beinhalten: Kompositionalität neuronaler Regelrepräsentationen im VLPFK; keine Evidenz für Regelreihenfolgenrepräsentation im VLPFK, welches gegen VLPFK als generelle Task-Set-Kontrollregion spricht; kein Hinweis auf einen hierarchieabhängigen Gradienten im VLPFK. Die komplementierende methodische Studie präsentiert "The Same Analysis Approach (SAA)", ein Ansatz zur Erkennung und Behebung experimentspezifischer Fehler, besonders solcher, die aus Design–Analyse–Interaktionen entstehen. SAA ist für relevant MVPA, aber auch für anderen Bereichen innerhalb und außerhalb der Neurowissenschaften.
Where and how does the brain represent complex rule sets? This thesis presents a series of three empirical studies that decompose representations of complex rule sets to directly address this question. An additional methodological study investigates the employed analysis method and the experimental design. The empirical studies employ multivariate pattern analysis (MVPA) of functional magnetic resonance imaging (fMRI) data from healthy human participants. The methodological study has been inspired by the empirical work. Its impact and application range, however, extend well beyond the empirical studies of this thesis. Questions of the empirical studies (Studies 1-3) include: Where are cues and rules represented, and are these represented independently? Where are compound rules (rules consisting of multiple rules) represented, and are these composed from their single rule representations? Where are rules from different hierarchical levels represented, and is there a hierarchy-dependent functional gradient along ventro-lateral prefrontal cortex (VLPFC)? Where is the order of rule-execution represented, and is it represented as a separate higher-level rule? All empirical studies employ information-based functional mapping ("searchlight" approach) to localise representations of rule set features brain-wide and spatially unbiased. Key findings include: compositional coding of compound rules in VLPFC; no order information in VLPFC, suggesting VLPFC is not a general controller for task set; evidence against the hypothesis of a hierarchy-dependent functional gradient along VLPFC. The methodological study (Study 4) introduces "The Same Analysis Approach (SAA)". SAA allows to detect, avoid, and eliminate confounds and other errors in experimental design and analysis, especially mistakes caused by malicious experiment-specific design-analysis interactions. SAA is relevant for MVPA, but can also be applied in other fields, both within and outside of neuroscience.

As we gain increased understanding of genetic disorders and gene regulation more focus has turned towards complex interactions. Combinations of genes or gene and environmental factors have been suggested to explain the missing heritability behind complex diseases. Furthermore, gene activation and splicing seem to be governed by a complex machinery of histone modification (HM), transcription factor (TF), and DNA sequence signals. This thesis aimed to apply and develop multivariate machine learning methods for use on such biological problems. Monte Carlo feature selection was combined with rule-based classification to identify interactions between HMs and to study the interplay of factors with importance for asthma and allergy. Firstly, publicly available ChIP-seq data (Paper I) for 38 HMs was studied. We trained a classifier for predicting exon inclusion levels based on the HMs signals. We identified HMs important for splicing and illustrated that splicing could be predicted from the HM patterns. Next, we applied a similar methodology on data from two large birth cohorts describing asthma and allergy in children (Paper II). We identified genetic and environmental factors with importance for allergic diseases which confirmed earlier results and found candidate gene-gene and gene-environment interactions. In order to interpret and present the classifiers we developed Ciruvis, a web-based tool for network visualization of classification rules (Paper III). We applied Ciruvis on classifiers trained on both simulated and real data and compared our tool to another methodology for interaction detection using classification. Finally, we continued the earlier study on epigenetics by analyzing HM and TF signals in genes with or without evidence of bidirectional transcription (Paper IV). We identified several HMs and TFs with different signals between unidirectional and bidirectional genes. Among these, the CTCF TF was shown to have a well-positioned peak 60-80 bp upstream of the transcription start site in unidirectional genes.

Paketklassifikation ist eine Kernfunktionalität vieler Netzwerksysteme, wie zum Beispiel Firewalls und SDN-Switches. Für viele dieser Systeme ist Durchsatz von höchster Bedeutung. Weitere wichtige Eigenschaften sind dynamische Aktualisierbarkeit und hohe Regelsatz-Ausdrucksfähigkeit. Die Kombination dieser Eigenschaften macht Paketklassifikation zu einem schwierigen Problem. Diese Arbeit befasst sich mit dem Design von Klassifikationssystemen und -algorithmen, welche mindestens zwei dieser Eigenschaften vereinen. Es werden hybride Systeme sowie Systemspezialisierung verwendet, um effiziente Ansätze zum Paketklassifikationsproblem in drei Bereichen zu erarbeiten: Klassifikationsalgorithmen, Regelsatztransformation und hardwarebasierte Architekturen. Die Beiträge im Bereich der Klassifikationsalgorithmen sind Jit Vector Search (JVS) und das SFL-System. JVS verbessert existierende Techniken durch spezialisierte Suchdatenstrukturen und durch Nutzung von SIMD-Fähigkeiten der CPU, was in fast optimaler Klassifikationsperformanz bei kaum erhöhten Vorberechnungszeiten resultiert. Das hybride SFL-System hingegen kombiniert einen Klassifikationsalgorithmus mit einem Änderungspuffer, um sowohl hohe Klassifikations- als auch Aktualisierungsperformanz zu ermöglichen. Bezüglich Regelsatztransformationen wird die RuleBender-Technik vorgestellt, welche Suchbäume in Regelsätze für Firewalls mit Sprungsemantik kodiert. Somit kann der Durchsatz dieser Systeme unter Beibehaltung komplexer Regelsatzsemantik um eine Größenordnung gesteigert werden. Schließlich wird der MPFC-Ansatz vorgestellt, welcher einen Regelsatz in einen auf einem FPGA implementierbaren Matching-Schaltkreis übersetzt. Die generierten Schaltkreise sind hochoptimiert und kleiner als generische Matching-Schaltkreise. Um dynamische Regelsatzänderungen zu ermöglichen, wird der hybride Consul-Ansatz konzipiert, welcher MPFC-Matcher mit generischen Matching-Schaltkreisen kombiniert.
Packet classification is a core functionality of a wide variety of network systems, such as firewalls and SDN switches. For many of these systems, throughput is of paramount importance. Further important system traits are dynamic updateability and high expressiveness in terms of rule set semantics. The combination of several of these properties turns packet classification into a hard problem. This work focuses on the design of classification systems and algorithms that combine at least two of the abovementioned characteristics. To this end, the concepts of hybrid systems and system specialization are employed to obtain efficient approaches to the packet classification problem in three domains: classification algorithms, rule set transformation, and hardware-centric architectures. The contributions in the domain of classification algorithms are Jit Vector Search (JVS) and the SFL system. JVS improves upon existing techniques through specialized search data structures and by exploiting SIMD capabilities of the underlying CPU, which results in near-optimal classification performance at only slightly increased preprocessing times. In contrast, the SFL system is a hybrid approach that combines a classification algorithm with an update buffer to allow for high classification as well as update performance. With respect to rule set transformation, the RuleBender technique is proposed, which encodes search tree structures into rule sets of firewalls with jump semantics. That way, the throughput of these systems can be improved by an order of magnitude, while maintaining complex matching semantics. Finally, the MPFC approach is proposed, which translates a given rule set into a matching circuit that can be implemented on an FPGA. The generated circuits are highly optimized and significantly smaller than those of generic matchers. To allow for dynamic rule set updates, the hybrid Consul approach is devised, which combines MPFC circuits with a generic matcher.

An expert system is useful when conventional programming techniques fall short of capturing human expert knowledge and making decisions using this information. In this study, we describe a framework for capturing expert knowledge under a decision tree form and this framework can be used for making decisions based on captured knowledge. The framework proposed in this study is generic and can be used to create domain specific expert systems for different problems. Features are created or processed by the nodes of decision tree and a final conclusion is reached for each feature. Framework supplies 3 types of nodes to construct a decision tree. First type is the decision node, which guides the search path with its answers. Second type is the operator node, which creates new features using the inputs. Last type of node is the end node, which corresponds to a conclusion about a feature. Once the nodes of the tree are developed, then user can interactively create the decision tree and run the supplied inference engine to collect the result on a specific problem. The framework proposed is experimented with two case studies
"
Airport Runway Detection in High Resolution Satellite Images"
and "
Urban Area Detection in High Resolution Satellite Images"
. In these studies linear features are used for structural decisions and Scale Invariant Feature Transform (SIFT) features are used for testing existence of man made structures.

The report presents an implemention of a classification algorithm for the Institutional Analysis Project. The algorithm used in this project is the decision tree classification algorithm which uses a gain ratio attribute selectionmethod. The algorithm discovers the hidden rules from the student records, which are used to predict whether or not other students are at risk of dropping out. It is shown that special rules exist in different data sets, each with their natural hidden knowledge. In other words, the rules that are obtained depend on the data that is used for classification. In our preliminary experiments, we show that between 55-78 percent of data with unknown class lables can be correctly classified, using the rules obtained from data whose class labels are known. We feel this is acceptable, given the large number of records, attributes, and attribute values that are used in the experiments. The project results are useful for large data set analysis.
viii, 38 leaves ; 29 cm. --

This volume discusses the phenomenon of ethnicity in East Africa under German colonial rule, using five case studies: the Swahili, Nyamwezi, Maasai, Shambaa and Bondei. Besides offering a brief overview of the debate concerning ethnicity in Africa and of the history of German colonial rule in East Africa the study examines the role played by missions in the development of ethnic consciousness. It also compares German and British colonial rule in this field.
Dieser Band setzt sich mit dem Phänomen der Ethnizität in Ostafrika unter deutscher Kolonialherrschaft auseinander, wobei fünf Fallstudien genutzt werden: die Swahili, Nyamwezi, Maasai, Shambaa und Bondei. Neben einem kurzen Überblick über die Debate bezüglich der Ethnizität in Afrika und der Geschichte der deutschen Kolonialherrschaft in Ostafrika, untersucht die Studie die Rolle der Missionen für die Entwicklung eines ethnischen Bewusstseins. Zu diesem Thema wird die deutsche Kolonialherrschaft auch mit der britischen verglichen.

The so-called 3σ-rule is a simple and widely used heuristic for outlier detection. This term is a generic term of some statistical hypothesis tests whose test statistics are known as normalized or studentized residuals. The conditions, under which this rule is statistically substantiated, were analyzed, and the extent it applies to geodetic least-squares adjustment was investigated. Then, the efficiency or non-efficiency of this method was analyzed and demonstrated on the example of repeated observations.
Die sogenannte 3σ-Regel ist eine einfache und weit verbreitete Heuristik für die Ausreißererkennung. Sie ist ein Oberbegriff für einige statistische Hypothesentests, deren Teststatistiken als normierte oder studentisierte Verbesserungen bezeichnet werden. Die Bedingungen, unter denen diese Regel statistisch begründet ist, werden analysiert. Es wird untersucht, inwieweit diese Regel auf geodätische Ausgleichungsprobleme anwendbar ist. Die Effizienz oder Nichteffizienz dieser Methode wird analysiert und demonstriert am Beispiel von Wiederholungsmessungen.

Contemporary technological systems generate massive quantities of log messages. These messages can be stored, searched and visualized efficiently using log management and analysis tools. The analysis of log messages offer insights into system behavior such as performance, server status and execution faults in web applications. iStone AB wants to explore the possibility to automate their debugging process. Since iStone does most parts of their debugging manually, it takes time to find errors within the system. The aim was therefore to find different solutions to reduce the time it takes to debug. An analysis of log messages within access – and console logs were made, so that the most appropriate data mining techniques for iStone’s system would be chosen. Data mining algorithms and log management and analysis tools were compared. The result of the comparisons showed that the ELK Stack as well as a mixture between Eclat and a hybrid algorithm (Eclat and Apriori) were the most appropriate choices. To demonstrate their feasibility, the ELK Stack and Eclat were implemented. The produced results show that data mining and the use of a platform for log analysis can facilitate and reduce the time it takes to debug.
Dagens system genererar stora mängder av loggmeddelanden. Dessa meddelanden kan effektivt lagras, sökas och visualiseras genom att använda sig av logghanteringsverktyg. Analys av loggmeddelanden ger insikt i systemets beteende såsom prestanda, serverstatus och exekveringsfel som kan uppkomma i webbapplikationer. iStone AB vill undersöka möjligheten att automatisera felsökning. Eftersom iStone till mestadels utför deras felsökning manuellt så tar det tid att hitta fel inom systemet. Syftet var att därför att finna olika lösningar som reducerar tiden det tar att felsöka. En analys av loggmeddelanden inom access – och konsolloggar utfördes för att välja de mest lämpade data mining tekniker för iStone’s system. Data mining algoritmer och logghanteringsverktyg jämfördes. Resultatet av jämförelserna visade att ELK Stacken samt en blandning av Eclat och en hybrid algoritm (Eclat och Apriori) var de lämpligaste valen. För att visa att så är fallet så implementerades ELK Stacken och Eclat. De framställda resultaten visar att data mining och användning av en plattform för logganalys kan underlätta och minska den tid det tar för att felsöka.

Machine learning approaches have been successfully applied to many classification and prediction problems. One of the most popular machine learning approaches is decision trees. A main advantage of decision trees is the clarity of the decision model they produce. The ID3 algorithm proposed by Quinlan forms the basis for many of the decision trees’ application. Trees produced by ID3 are sensitive to small perturbations in training data. To overcome this problem and to handle data uncertainties and spurious precision in data, fuzzy ID3 integrated fuzzy set theory and ideas from fuzzy logic with ID3. Several fuzzy decision trees algorithms and tools exist. However, existing tools are slow, produce a large number of rules and/or lack the support for automatic fuzzification of input data. These limitations make those tools unsuitable for a variety of applications including those with many features and real time ones such as intrusion detection. In addition, the large number of rules produced by these tools renders the generated decision model un-interpretable. In this research work, we proposed an improved version of the fuzzy ID3 algorithm. We also introduced a new method for reducing the number of fuzzy rules generated by Fuzzy ID3. In addition we applied fuzzy decision trees to the classification of real and pseudo microRNA precursors. Our experimental results showed that our improved fuzzy ID3 can achieve better classification accuracy and is more efficient than the original fuzzy ID3 algorithm, and that fuzzy decision trees can outperform several existing machine learning algorithms on a wide variety of datasets. In addition our experiments showed that our developed fuzzy rule reduction method resulted in a significant reduction in the number of produced rules, consequently, improving the produced decision model comprehensibility and reducing the fuzzy decision tree execution time. This reduction in the number of rules was accompanied with a slight improvement in the classification accuracy of the resulting fuzzy decision tree. In addition, when applied to the microRNA prediction problem, fuzzy decision tree achieved better results than other machine learning approaches applied to the same problem including Random Forest, C4.5, SVM and Knn.

Geographic object-based image analysis (GEOBIA) is an innovative image classification technique that treats spatial features in an image as objects, rather than as pixels; thus resembling closer to that of human perception of the geographic space. However, the process of a GEOBIA application allows for multiple interpretations. Particularly sensitive parts of the process include image segmentation and training data selection. The multiresolution segmentation algorithm (MSA) is commonly applied. The performance of segmentation depends primarily on the algorithms scale parameter, since scale controls the size of image objects produced. The fact that the scale parameter is unit less makes it a challenge to select a suitable one; thus, leaving the analyst to a method of trial and error. This can lead to a possible bias. Additionally, part from the segmentation, training area selection usually means that the data has to be manually collected. This is not only time consuming but also prone to subjectivity. In order to overcome these challenges, we tested a GEOBIA scheme that involved automatic methods of MSA scale parameterisation and training area selection which enabled us to more objectively classify images. Three study areas within Sweden were selected. The data used was high resolution Geografiska Sverigedata (GSD) orthophotos from the Swedish mapping agency, Lantmäteriet. We objectively found scale for each classification using a previously published technique embedded as a tool in eCognition software. Based on the orthophoto inputs, the tool calculated local variance and rate of change at different scales. These figures helped us to determine scale value for the MSA segmentation. Moreover, we developed in this study a novel method for automatic training area selection. The method is based on thresholded feature statistics layers computed from the orthophoto band derivatives. Thresholds were detected by Otsu’s single and multilevel algorithms. The layers were run through a filtering process which left only those fit for use in the classification process. We also tested the transferability of classification rule-sets for two of the study areas. This test helped us to investigate the degree to which automation can be realised. In this study we have made progress toward a more objective way of object-based image classification, realised by automating the scheme. Particularly noteworthy is the algorithm for automatic training area selection proposed, which compared to manual selection restricts human intervention to a minimum. Results of the classification show overall well delineated classes, in particular, the border between open area and forest contributed by the elevation data. On the other hand, there still persists some challenges regarding separating between deciduous and coniferous forest. Furthermore, although water was accurately classified in most instances, in one of the study areas, the water class showed contradictory results between its thematic and positional accuracy; hence stressing the importance of assessing the result based on more than the thematic accuracy. From the transferability test we noted the importance of considering the spatial/spectral characteristics of an area before transferring of rule-sets as these factors are a key to determine whether a transfer is possible.

In this thesis, we discuss different techniques to bridge the gap between two different approaches to artificial intelligence: the symbolic and the connectionist paradigm. Both approaches have quite contrasting advantages and disadvantages. Research in the area of neural-symbolic integration aims at bridging the gap between them. Starting from a human readable logic program, we construct connectionist systems, which behave equivalently. Afterwards, those systems can be trained, and later the refined knowledge be extracted.

Cette thèse explique et présente notre démarche de la réalisation d’un système à base de règles de reconnaissance et de classification automatique des EN en arabe. C’est un travail qui implique deux disciplines : la linguistique et l’informatique. L’outil informatique et les règles la linguistiques s’accouplent pour donner naissance à une nouvelle discipline ; celle de « traitement automatique des langues », qui opère sur des niveaux différents (morphosyntaxique, syntaxique, sémantique, syntactico-sémantique etc.). Nous avons donc, dans ce qui nous concerne, mis en œuvre des informations et règles linguistiques nécessaires au service du logiciel informatique, qui doit être en mesure de les appliquer, pour extraire et classifier, par des annotations syntaxiques et/ou sémantiques, les différentes classes d’entités nommées.Ce travail de thèse s’inscrit donc dans un cadre général de traitement automatique des langues, mais plus particulièrement dans la continuité des travaux réalisés au niveau de l’analyse morphosyntaxique par la conception et la réalisation des bases des données lexicales SAMIA et ensuite DIINAR avec l’ensemble de résultats de recherches qui en découlent. C’est une tâche qui vise à l’enrichissement lexical par des entités nommées simples et complexes, et qui veut établir la transition de l’analyse morphosyntaxique vers l’analyse syntaxique, et syntatico-sémantique dans une visée plus générale de l’analyse du contenu textuel. Pour comprendre de quoi il s’agit, il nous était important de commencer par la définition de l’entité nommée. Et pour mener à bien notre démarche, nous avons distingué entre deux types principaux : pur nom propre et EN descriptive. Nous avons aussi établi une classification référentielle en se basant sur diverses classes et sous-classes qui constituent la référence de nos annotations sémantiques. Cependant, nous avons dû faire face à deux difficultés majeures : l’ambiguïté lexicale et les frontières des entités nommées complexes. Notre système adopte une approche à base de règles syntactico-sémantiques. Il est constitué, après le Niveau 0 d’analyse morphosyntaxique, de cinq niveaux de construction de patrons syntaxiques et syntactico-sémantiques basés sur les informations linguistique nécessaires (morphosyntaxiques, syntaxiques, sémantique, et syntactico-sémantique). Ce travail, après évaluation en utilisant deux corpus, a abouti à de très bons résultats en précision, en rappel et en F–mesure. Les résultats de notre système ont un apport intéressant dans différents application du traitement automatique des langues notamment les deux tâches de recherche et d’extraction d’informations. En effet, on les a concrètement exploités dans les deux applications (recherche et extraction d’informations). En plus de cette expérience unique, nous envisageons par la suite étendre notre système à l’extraction et la classification des phrases dans lesquelles, les entités classifiées, principalement les entités nommées et les verbes, jouent respectivement le rôle d’arguments et de prédicats. Un deuxième objectif consiste à l’enrichissement des différents types de ressources lexicales à l’instar des ontologies
This thesis explains and presents our approach of rule-based system of arabic named entity recognition and classification. This work involves two disciplines : linguistics and computer science. Computer tools and linguistic rules are merged to give birth to a new discipline : Natural Languge Processsing, which operates in different levels (morphosyntactic, syntactic, semantic, syntactico-semantic…). So, in our particular case, we have put the necessary linguistic information and rules to software sevice. This later should be able to apply and implement them in order to recognise and classify, by syntactic and semantic annotations, the different named entity classes.This work of thesis is incorporated within the general domain of natural language processing, but it particularly falls within the scope of the continuity of the accomplished work in terms of morphosyntactic analysis and the realisation of lexical data bases of SAMIA and then DIINAR as well as the accompanying scientific recearch. This task aimes at lexical enrichement with simple and complex named entities and at establishing the transition from the morphological analysis into syntactic and syntactico-semantic analysis. The ultimate objective is text analysis. To understand what it is about, it was important to start with named entity definition. To carry out this task, we distinguished between two main named entity types : pur proper name and descriptive named entities. We have also established a referential classification on the basis of different classes and sub-classes which constitue the reference for our semantic annotations. Nevertheless, we are confronted with two major difficulties : lexical ambiguity and the frontiers of complex named entities. Our system adoptes a syntactico-semantic rule-based approach. After Level 0 of morpho-syntactic analysis, the system is made up of five levels of syntactic and syntactico-semantic patterns based on tne necessary linguisic information (i.e. morphosyntactic, syntactic, semantic and syntactico-semantic information).This work has obtained very good results in termes of precision, recall and F-measure. The output of our system has an interesting contribution in different applications of the natural language processing especially in both tasks of information retrieval and information extraction. In fact, we have concretely exploited our system output in both applications (information retrieval and information extraction). In addition to this unique experience, we envisage in the future work to extend our system into the sentence extraction and classification, in which classified entities, mainly named entities and verbs, play respectively the role of arguments and predicates. The second objective consists in the enrichment of different types of lexical resources such as ontologies

Below vegetation, throughfall kinetic energy (TKE) is an important factor to express the potential of rainfall to detach soil particles and thus for predicting soil erosion rates. TKE is affected by many biotic (e.g. tree height, leaf area index) and abiotic (e.g. throughfall amount) factors because of changes in rain drop size and velocity. However, studies modelling TKE with a high number of those factors are lacking. This study presents a new approach to model TKE. We used 20 biotic and abiotic factors to evaluate thresholds of those factors that can mitigate TKE and thus decrease soil erosion. Using these thresholds, an optimal set of biotic and abiotic factors was identified to minimize TKE. The model approach combined recursive feature elimination, random forest (RF) variable importance and classification and regression trees (CARTs). TKE was determined using 1405 splash cup measurements during five rainfall events in a subtropical Chinese tree plantation with five-year-old trees in 2013. Our results showed that leaf area, tree height, leaf area index and crown area are the most prominent vegetation traits to model TKE. To reduce TKE, the optimal set of biotic and abiotic factors was a leaf area lower than 6700mm2, a tree height lower than 290 cm combined with a crown base height lower than 60 cm, a leaf area index smaller than 1, more than 47 branches per tree and using single tree species neighbourhoods. Rainfall characteristics, such as amount and duration, further classified high or low TKE. These findings are important for the establishment of forest plantations that aim to minimize soil erosion in young succession stages using TKE modelling.

This volume discusses the history of funerals, a 'total social phenomenon' in southern Ghana. Today, as in the past, festivals are organised, usually involving music, dance and the consumption of alcohol. This study discusses variations over time and between different regions, dealing systematically with the preparation of the corpse, places of burial, modes of commemoration, the high costs involved and the influence of Christian missions.
Dieser Band betrachtet die Geschichte von Beerdigungen, ein 'total social phenomenon' im südlichen Ghana. Heute, wie auch in der Vergangenheit, werden Feste organisiert, die normalerweise Musik, Tanz und den Konsum von Alkohol involvieren. Diese Studie betrachtet Variationen über die Zeit hinweg und zwischen verschiedenen Regionen, während sie sich systematisch mit der Vorbereitung der Leiche, den Orten der Beerdigung, den Arten der Gedenkfeiern, den hohen Kosten und dem Einfluss der christlichen Missionen beschäftigt.