Дисертації з теми "Tree of decisions"

Brazilian productive forest plantations and sustainability is a continuous subject of debate, since such forest plantations constitute an increasing proportion of the Brazilian forest cover, and because of documented instances where some establishments have resulted in negative impacts on local ecosystems and communities. Meanwhile, balancing economic, environmental and social sustainability is becoming an important concern in industrial decision-making given the increasing importance of the global sustainability goals. The aim of this study is to investigate the contemporary decision context in which companies in the Brazilian tree industry select locations for forest plantations in Brazil. Special focus is devoted to how sustainability aspects are included in such decisions, as well as the roles assigned to possible mechanisms for sustainable development, such as Brazilian policy, legislation and independent forest certification. A qualitative research strategy is employed, which encompasses a literature review and semi-structured interviews with industry practitioners in two segments of the industry, whose perceptions are triangulated with perspectives of relevant non-industry actors. Grounded theory is used to analyse the qualitative data. The findings from the literature and the qualitative data are synthesised, and several hypotheses are consequently developed regarding forest plantation location decisions and sustainability in Brazil. The hypotheses evolve around the finding that strategic, economic, environmental and social perspectives are perceived as present in contemporary decision-making, in certified pulp and paper companies. Moreover, a hypothesis of general character is developed based on the synthesised findings, which is that forest plantations can be the subject of integrated location and sustainability analyses, where the problem can be formulated as finding an optimal location for a forest plantation from a three-dimensional sustainability perspective. Finally, some prospects for further integrated research about forest plantation location and sustainability are presented.
Denna studie handlar om brasilianska skogsplanteringar och expansionen av planteringar som ägs av företag i landets, så kallade ”trädindustri”. Studien fokuserar på produktiva skogsplanteringar av introducerade arter, för vilka intresset har vuxit parallellt med att de har expanderat till yta under de senaste decennierna. Intresset har även vuxit eftersom federala beslutsfattare har lyft fram skogsplanteringar som ett verktyg för att minska landets koldioxidutsläpp och stimulera ekonomisk utveckling. Produktiva brasilianska skogsplanteringar och hållbarhet är ett kontinuerligt föremål för debatt. Den befintliga litteraturen om expansionen av sådana planteringar indikerar att företag i landets trädindustri inte alltid har balanserat de ekonomiska, miljömässiga och sociala hållbarhetsdimensionerna i sina beslut gällande var planteringar ska anläggas. I vissa fall har besluten resulterat i negativa följder för lokala ekosystem och samhällen. Vidare finns det, såvitt författaren vet, inga studier som behandlar lokaliseringsbeslut gällande sådana brasilianska skogsplanteringar och alla de tre hållbarhetsdimensionerna. Därför syftar denna studie till att förbättra förståelsen för det samtida beslutssammanhang i vilket företag i den brasilianska trädindustrin väljer platser för sina skogsplanteringar, samt hur hållbarhetsaspekter omfattas i sådana beslut. Som ett led i denna strävan undersöks möjliga platsfaktorer, rumsliga begränsningar och andra beslutsaspekter, liksom vilken roll företag tilldelar specifika mekanismer för hållbar utveckling, såsom brasiliansk policy, lagstiftning och oberoende skogscertifiering. En kvalitativ forskningsstrategi antas för att genomföra undersökningen. En litteraturöversikt kombineras med semistrukturerade intervjuer med branschutövare i två segment av den brasilianska trädindustrin. Deras uppfattningar trianguleras med perspektiv från relevanta aktörer utanför branschen. Urvalet av deltagarna för studien genomfördes på plats i Brasilien 2015 till 2016 och resulterade i 13 brasilianska intervjudeltagare. Intervjuerna genomfördes sedan på olika platser i Brasilien under 2016. Grundad teori används som forskningsmetod för att analysera insamlad kvalitativ data. Två huvudteman, flera sekundära teman och ett koncept härleds från de kvalitativa data som samlades in, vilka främst grundar sig i uppfattningar gällande brasilianska massa- och pappersföretag. Flera beslutsfaktorer och begränsningar som kan påverka placeringen av brasilianska skogsplanteringar identifieras också. Det kan konstateras att aktörerna i urvalet i studien uppfattar det som att strategiska, ekonomiska, miljömässiga och sociala perspektiv är närvarande i dagens lokaliseringsbeslut, utförda av certifierade, brasilianska massa- och pappersföretag. En viktig begränsning i studien är att dess utforskande karaktär hindrar författaren från att presentera några ”sanningar” om ämnet som undersöks, eller att dra slutsatser om företagens verkliga handlingar. Därför presenteras istället ett antal hypoteser som gäller Brasilien, men även en hypotes av generell karaktär. Den generella hypotesen är att skogsplanteringar kan vara föremål för integrerad lokaliserings- och hållbarhetsanalys, där problemet kan formuleras som att hitta optimala platser för skogsplanteringar ur ett tredimensionellt hållbarhetsperspektiv. Med tanke på begränsningarna i studien, samt associerade osäkerheter som hypoteserna gällande det brasilianska fallet är befästa med, är rekommendationen att fortsatta studier först koncentrerar sig på att testa den allmänna hypotesen. Om fortsatta studier på nationell nivå visar sig vara fördelaktiga efter sådana tester, uppmuntras forskare att återvända till sammanfattningen av branschperspektiv, återstående frågor och hypoteser som tillgängliggörs för fortsatt forskning om Brasilien genom denna studie.
Este estudo trata das florestas plantadas brasileiras, e da expansão de plantios pertencentes a empresas da indústria brasileira de árvores plantadas. O estudo tem foco nas florestas plantadas produtivas de espécies introduzidas, pelas quais se teve um aumento no interesse paralelamente à sua expansão geográfica nas últimas décadas. O interesse também aumentou ao destaque dado aos plantios florestais pelo governo federal como uma ferramenta para reduzir as emissões de dióxido de carbono do país, e estimular o desenvolvimento econômico. As florestas plantadas produtivas brasileiras e a sustentabilidade são constantemente temas de debate. A literatura existente sobre essa expansão indica que as empresas da indústria brasileira de árvores plantadas nem sempre equilibram as três dimensões de sustentabilidade ao decidir onde plantar suas florestas. Em alguns casos as decisões resultam em impactos negativos nos ecossistemas e comunidades locais. Além disso, com base no conhecimento da autora, não existem estudos que tratem das decisões de localização das florestas plantadas brasileiras e de todas essas três dimensões da sustentabilidade. Por tanto, este estudo tem como objetivo melhorar a compreensão do contexto atual de tomada de decisões em que as empresas na indústria brasileira de árvores plantadas escolhem os locais para suas florestas plantadas, e como os aspectos de sustentabilidade são incluídos em tais decisões. Como parte desse empenho, são examinadas possíveis limitações e outros aspectos de tomada de decisão, bem como o papel que as empresas atribuem à certos mecanismos para o desenvolvimento sustentável, como a política brasileira, a legislação e a certificação florestal. Uma estratégia de pesquisa qualitativa é adotada para conduzir a pesquisa. Uma revisão de literatura é combinada com entrevistas semiestruturadas com profissionais em dois segmentos da indústria brasileira de árvores plantadas. Suas percepções são trianguladas com perspectivas de atores relevantes de fora da indústria. A amostra de participantes do estudo foi realizada no Brasil entre 2015 e 2016 e resultou em 13 participantes. As entrevistas foram então realizadas em diversos locais no Brasil em 2016. É utilizada a teoria fundamentada nos dados como método de pesquisa para analisar os dados qualitativos coletados. São extraídos dois temas principais, diversos temas secundários e um conceito a partir dos dados qualitativos coletados, baseados principalmente nas percepções das empresas brasileiras de papel e celulose. Vários fatores de decisão que podem influenciar a localização das plantações florestais também são identificados. Observa-se que os participantes da amostra do estudo percebem que perspectivas estratégicas, econômicas, ambientais e sociais estão presentes nas decisões atuais de localização, realizadas por empresas certificadas de celulose e papel. Uma importante limitação do estudo é que sua característica exploratória impede que a pesquisadora apresente “verdades” sobre o assunto investigado, ou tire conclusões sobre os atos das empresas. Portanto, são apresentadas várias hipóteses aplicáveis ao Brasil, mas também uma hipótese de caráter geral. A hipótese geral é que as plantações florestais podem estar sujeitas a análises integradas de localização e sustentabilidade, onde o problema pode ser formulado como encontrar um local ideal para uma plantação florestal a partir de uma perspectiva tridimensional de sustentabilidade. Dadas as limitações do estudo, bem como as incertezas associadas às quais as hipóteses do Brasil estão relacionadas, a recomendação é que novos estudos se concentrem primeiro em testar a hipótese geral. Caso novos estudos em nível nacional forem benéficos após esses testes, os pesquisadores são incentivados a retornar ao resumo das perspectivas da indústria, das questões remanescentes e das hipóteses disponibilizadas para futuras pesquisas sobre o Brasil por meio deste estudo.

"SELECT CD is a site-specific, decision-support tool for selecting ecologically viable tree species, reproduction cuttings, and regeneration methods in the coastal forest (CDF, CWH, and MH zones). SELECT CD integrates information from several existing guides with new information from literature and recent research into a single, user-friendly resource. SELECT CD also includes a rich library of visuals and an illustrated glossary of technical terms."

The thesis proposes novel full decision tree and decision tree ensembleinduction algorithms EFTI and EEFTI, and various possibilities for theirimplementations are explored. The experiments show that the proposed EFTIalgorithm is able to infer much smaller DTs on average, without thesignificant loss in accuracy, when compared to the top-down incremental DTinducers. On the other hand, when compared to other full tree inductionalgorithms, it was able to produce more accurate DTs, with similar sizes, inshorter times. Also, the hardware architectures for acceleration of thesealgorithms (EFTIP and EEFTIP) are proposed and it is shown in experimentsthat they can offer substantial speedups.
У овоj дисертациjи, представљени су нови алгоритми EFTI и EEFTI заформирање стабала одлуке и њихових ансамбала неинкременталномметодом, као и разне могућности за њихову имплементациjу.Експерименти показуjу да jе предложени EFTI алгоритам у могућностида произведе драстично мања стабла без губитка тачности у односу напостојеће top-down инкременталне алгоритме, а стабла знатно већетачности у односу на постојеће неинкременталне алгоритме. Такође супредложене хардверске архитектуре за акцелерацију ових алгоритама(EFTIP и EEFTIP) и показано је да је уз помоћ ових архитектура могућеостварити знатна убрзања.
U ovoj disertaciji, predstavljeni su novi algoritmi EFTI i EEFTI zaformiranje stabala odluke i njihovih ansambala neinkrementalnommetodom, kao i razne mogućnosti za njihovu implementaciju.Eksperimenti pokazuju da je predloženi EFTI algoritam u mogućnostida proizvede drastično manja stabla bez gubitka tačnosti u odnosu napostojeće top-down inkrementalne algoritme, a stabla znatno većetačnosti u odnosu na postojeće neinkrementalne algoritme. Takođe supredložene hardverske arhitekture za akceleraciju ovih algoritama(EFTIP i EEFTIP) i pokazano je da je uz pomoć ovih arhitektura mogućeostvariti znatna ubrzanja.

Decision trees (DTs) play a vital role in statistical modelling. Simplicity and interpretability of the solution structure have made the method popular in a wide range of disciplines. In data classification problems, DTs recursively partition the feature space into disjoint sub-regions until each sub-region becomes homogeneous with respect to a particular class. Axis parallel splits, the simplest form of splits, partition the feature space parallel to feature axes. However, for some problem domains DTs with axis parallel splits can produce complicated boundary structures. As an alternative, oblique splits are used to partition the feature space potentially simplifying the boundary structure. Various approaches have been explored to find optimal oblique splits. One approach is based on optimisation techniques. This is considered the benchmark approach, however, its major limitation is that the tree induction algorithm is computationally expensive. On the other hand, split finding approaches based on heuristic arguments have gained popularity and have made improvements on benchmark methods. This thesis proposes a methodology to induce oblique decision trees in transformed spaces based on a heuristic argument. As the first goal of the thesis, a new oblique decision tree algorithm, called HHCART (\underline{H}ouse\underline{H}older \underline{C}lassification and \underline{R}egression \underline{T}ree) is proposed. The proposed algorithm utilises a series of Householder matrices to reflect the training data at each non-terminal node during the tree construction. Householder matrices are constructed using the eigenvectors from each classes' covariance matrix. Axis parallel splits in the reflected (or transformed) spaces provide an efficient way of finding oblique splits in the original space. Experimental results show that the accuracy and size of the HHCART trees are comparable with some benchmark methods in the literature. The appealing features of HHCART is that it can handle both qualitative and quantitative features in the same oblique split, conceptually simple and computationally efficient. Data mining applications often come with massive example sets and inducing oblique DTs for such example sets often consumes considerable time. HHCART is a serial computing memory resident algorithm which may be ineffective when handling massive example sets. As the second goal of the thesis parallel computing and disk resident versions of the HHCART algorithm are presented so that HHCART can be used irrespective of the size of the problem. HHCART is a flexible algorithm and the eigenvectors defining Householder matrices can be replaced by other vectors deemed effective in oblique split finding. The third endeavour of this thesis explores this aspect of HHCART. HHCART can be used with other vectors in order to improve classification results. For example, a normal vector of the angular bisector, introduced in the Geometric Decision Tree (GDT) algorithm, is used to construct the Householder reflection matrix. The proposed method produces better results than GDT for some problem domains. In the second case, \textit{Class Representative Vectors} are introduced and used to construct Householder reflection matrices. The results of this experiment show that these oblique trees produce classification results competitive with those achieved with some benchmark decision trees. DTs are constructed using two approaches, namely: top-down and bottom-up. HHCART is a top-down tree, which is the most common approach. As the fourth idea of the thesis, the concept of HHCART is used to induce a new DT, HHBUT, using the bottom-up approach. The bottom-up approach performs cluster analysis prior to the tree building to identify the terminal nodes. The use of the Bayesian Information Criterion (BIC) to determine the number of clusters leads to accurate and compact trees when compared with Cross Validation (CV) based bottom-up trees. We suggest that HHBUT is a good alternative to the existing bottom-up tree especially when the number of examples is much higher than the number of features.

In best-first top-down induction of decision trees, the best split is added in each step (e.g. the split that maximally reduces the Gini index). This is in contrast to the standard depth-first traversal of a tree. The resulting tree will be the same, just how it is built is different. The objective of this project is to investigate whether it is possible to determine an appropriate tree size on practical datasets by combining best-first decision tree growth with cross-validation-based selection of the number of expansions that are performed. Pre-pruning, post-pruning, CART-pruning can be performed this way to compare.

A hyper-heuristic is any algorithm that searches or operates in the space of heuristics as opposed to the space of solutions. Hyper-heuristics are increasingly used in function and combinatorial optimization. Rather than attempt to solve a problem using a fixed heuristic, a hyper-heuristic approach attempts to find a combination of heuristics that solve a problem (and in turn may be directly suitable for a class of problem instances). Hyper-heuristics have been little explored in data mining. This work presents novel hyper-heuristic approaches to data mining, by searching a space of attribute selection criteria for decision tree building algorithm. The search is conducted by a genetic algorithm. The result of the hyper-heuristic search in this case is a strategy for selecting attributes while building decision trees. Most hyper-heuristics work by trying to adapt the heuristic to the state of the problem being solved. Our hyper-heuristic is no different. It employs a strategy for adapting the heuristic used to build decision tree nodes according to some set of features of the training set it is working on. We introduce, explore and evaluate five different ways in which this problem state can be represented for a hyper-heuristic that operates within a decisiontree building algorithm. In each case, the hyper-heuristic is guided by a rule set that tries to map features of the data set to be split by the decision tree building algorithm to a heuristic to be used for splitting the same data set. We also explore and evaluate three different sets of low-level heuristics that could be employed by such a hyper-heuristic. This work also makes a distinction between specialist hyper-heuristics and generalist hyper-heuristics. The main difference between these two hyperheuristcs is the number of training sets used by the hyper-heuristic genetic algorithm. Specialist hyper-heuristics are created using a single data set from a particular domain for evolving the hyper-heurisic rule set. Such algorithms are expected to outperform standard algorithms on the kind of data set used by the hyper-heuristic genetic algorithm. Generalist hyper-heuristics are trained on multiple data sets from different domains and are expected to deliver a robust and competitive performance over these data sets when compared to standard algorithms. We evaluate both approaches for each kind of hyper-heuristic presented in this thesis. We use both real data sets as well as synthetic data sets. Our results suggest that none of the hyper-heuristics presented in this work are suited for specialization – in most cases, the hyper-heuristic’s performance on the data set it was specialized for was not significantly better than that of the best performing standard algorithm. On the other hand, the generalist hyper-heuristics delivered results that were very competitive to the best standard methods. In some cases we even achieved a significantly better overall performance than all of the standard methods.

Advances in data collection methods, storage and processing technology are providing a unique challenge and opportunity for automated data learning techniques which aim at producing high-level information, or models, from data. A Typical knowledge discovery process consists of data selection, data preparation, data transformation, data mining and interpretation/validation of the results. Thus, we develop automatic learning techniques which contribute to the data preparation, transformation and mining tasks of knowledge discovery. In doing so, we try to improve the prediction accuracy of the overall learning process. Our work focuses on decision tree based learning and thus, we introduce various preprocessing and transformation techniques such as discretization, fuzzy partitioning and dimensionality reduction to improve this type of learning. However, these techniques can be used in other learning methods e.g. discretization can also be used for naive-bayes classifiers. The data preparation step represents almost 80 percent of the problem and is both time consuming and critical for the quality of modeling. Discretization of continuous features is an important problem that has effects on accuracy, complexity, variance and understandability of the induction models. In this thesis, we propose and develop resampling based aggregation techniques that improve the quality of discretization. Later, we validate by comparing with other discretization techniques and with an optimal partitioning method on 10 benchmark data sets.The second part of our thesis concerns with automatic fuzzy partitioning for soft decision tree induction. Soft or fuzzy decision tree is an extension of the classical crisp tree induction such that fuzzy logic is embedded into the induction process with the effect of more accurate models and reduced variance, but still interpretable and autonomous. We modify the above resampling based partitioning method to generate fuzzy partitions. In addition we propose, develop and validate another fuzzy partitioning method that improves the accuracy of the decision tree.Finally, we adopt a topological learning scheme and perform non-linear dimensionality reduction. We modify an existing manifold learning based technique and see whether it can enhance the predictive power and interpretability of classification
La recherche avancée dans les méthodes d'acquisition de données ainsi que les méthodes de stockage et les technologies d'apprentissage, s'attaquent défi d'automatiser de manière systématique les techniques d'apprentissage de données en vue d'extraire des connaissances valides et utilisables.La procédure de découverte de connaissances s'effectue selon les étapes suivants: la sélection des données, la préparation de ces données, leurs transformation, le fouille de données et finalement l'interprétation et validation des résultats trouvés. Dans ce travail de thèse, nous avons développé des techniques qui contribuent à la préparation et la transformation des données ainsi qu'a des méthodes de fouille des données pour extraire les connaissances. A travers ces travaux, on a essayé d'améliorer l'exactitude de la prédiction durant tout le processus d'apprentissage. Les travaux de cette thèse se basent sur les arbres de décision. On a alors introduit plusieurs approches de prétraitement et des techniques de transformation; comme le discrétisation, le partitionnement flou et la réduction des dimensions afin d'améliorer les performances des arbres de décision. Cependant, ces techniques peuvent être utilisées dans d'autres méthodes d'apprentissage comme la discrétisation qui peut être utilisées pour la classification bayesienne.Dans le processus de fouille de données, la phase de préparation de données occupe généralement 80 percent du temps. En autre, elle est critique pour la qualité de la modélisation. La discrétisation des attributs continus demeure ainsi un problème très important qui affecte la précision, la complexité, la variance et la compréhension des modèles d'induction. Dans cette thèse, nous avons proposes et développé des techniques qui ce basent sur le ré-échantillonnage. Nous avons également étudié d'autres alternatives comme le partitionnement flou pour une induction floue des arbres de décision. Ainsi la logique floue est incorporée dans le processus d'induction pour augmenter la précision des modèles et réduire la variance, en maintenant l'interprétabilité.Finalement, nous adoptons un schéma d'apprentissage topologique qui vise à effectuer une réduction de dimensions non-linéaire. Nous modifions une technique d'apprentissage à base de variété topologiques `manifolds' pour savoir si on peut augmenter la précision et l'interprétabilité de la classification

Design of ensemble classifiers involves three factors: 1) a learning algorithm to produce a classifier (base classifier), 2) an ensemble method to generate diverse classifiers, and 3) a combining method to combine decisions made by base classifiers. With regard to the first factor, a good choice for constructing a classifier is a decision tree learning algorithm. However, a possible problem with this learning algorithm is its complexity which has only been addressed previously in the context of pruning methods for individual trees. Furthermore, the ensemble method may require the learning algorithm to produce a complex classifier. Considering the fact that performance of simplification methods as well as ensemble methods changes from one domain to another, our main contribution is to address a simplification method (post-pruning) in the context of ensemble methods including Bagging, Boosting and Error-Correcting Output Code (ECOC). Using a statistical test, the performance of ensembles made by Bagging, Boosting and ECOC as well as five pruning methods in the context of ensembles is compared. In addition to the implementation a supporting theory called Margin, is discussed and the relationship of Pruning to bias and variance is explained. For ECOC, the effect of parameters such as code length and size of training set on performance of Pruning methods is also studied. Decomposition methods such as ECOC are considered as a solution to reduce complexity of multi-class problems in many real problems such as face recognition. Focusing on the decomposition methods, AdaBoost.OC which is a combination of Boosting and ECOC is compared with the pseudo-loss based version of Boosting, AdaBoost.M2. In addition, the influence of pruning on the performance of ensembles is studied. Motivated by the result that both pruned and unpruned ensembles made by AdaBoost.OC have similar accuracy, pruned ensembles are compared with ensembles of single node decision trees. This results in the hypothesis that ensembles of simple classifiers may give better performance as shown for AdaBoost.OC on the identification problem in face recognition. The implication is that in some problems to achieve best accuracy of an ensemble, it is necessary to select base classifier complexity.

One of the major tasks in Data Mining is classification. The growing of Decision Tree from data is a very efficient technique for learning classifiers. The selection of an attribute used to split the data set at each Decision Tree node is fundamental to properly classify objects; a good selection will improve the accuracy of the classification. In this paper, we study the behavior of the Decision Trees induced with 14 attribute selection measures over three data sets taken from UCI Machine Learning Repository.

This thesis develops a new approach to fault tree analysis, namely the Binary Decision Diagram (BDD) method. Conventional qualitative fault tree analysis techniques such as the "top-down" or "bottom-up" approaches are now so well developed that further refinement is unlikely to result in vast improvements in terms of their computational capability. The BDD method has exhibited potential gains to be made in terms of speed and efficiency in determining the minimal cut sets. Further, the nature of the binary decision diagram is such that it is more suited to Boolean manipulation. The BDD method has been programmed and successfully applied to a number of benchmark fault trees. The analysis capabilities of the technique have been extended such that all quantitative fault tree top event parameters, which can be determined by conventional Kinetic Tree Theory, can now be derived directly from the BDD. Parameters such as the top event probability, frequency of occurrence and expected number of occurrences can be calculated exactly using this method, removing the need for the approximations previously required. Thus the BDD method is proven to have advantages in terms of both accuracy and efficiency. Initiator/enabler event analysis and importance measures have been incorporated to extend this method into a full analysis procedure.

Cost-sensitive multiclass classification problems, in which the task of assessing the impact of the costs associated with different misclassification errors, continues to be one of the major challenging areas for data mining and machine learning. The literature reviews in this area show that most of the cost-sensitive algorithms that have been developed during the last decade were developed to solve binary classification problems where an example from the dataset will be classified into only one of two available classes. Much of the research on cost-sensitive learning has focused on inducing decision trees, which are one of the most common and widely used classification methods, due to the simplicity of constructing them, their transparency and comprehensibility. A review of the literature shows that inducing nonlinear multiclass cost-sensitive decision trees is still in its early stages and further research could result in improvements over the current state of the art. Hence, this research aims to address the following question: 'How can non-linear regions be identified for multiclass problems and utilized to construct decision trees so as to maximize the accuracy of classification, and minimize misclassification costs?' This research addresses this problem by developing a new algorithm called the Elliptical Cost-Sensitive Decision Tree algorithm (ECSDT) that induces cost-sensitive non-linear (elliptical) decision trees for multiclass classification problems using evolutionary optimization methods such as particle swarm optimization (PSO) and Genetic Algorithms (GAs). In this research, ellipses are used as non-linear separators, because of their simplicity and flexibility in drawing non-linear boundaries by modifying and adjusting their size, location and rotation towards achieving optimal results. The new algorithm was developed, tested, and evaluated in three different settings, each with a different objective function. The first considered maximizing the accuracy of classification only; the second focused on minimizing misclassification costs only, while the third considered both accuracy and misclassification cost together. ECSDT was applied to fourteen different binary-class and multiclass data sets and the results have been compared with those obtained by applying some common algorithms from Weka to the same datasets such as J48, NBTree, MetaCost, and the CostSensitiveClassifier. The primary contribution of this research is the development of a new algorithm that shows the benefits of utilizing elliptical boundaries for cost-sensitive decision tree learning. The new algorithm is capable of handling multiclass problems and an empirical evaluation shows good results. More specifically, when considering accuracy only, ECSDT performs better in terms of maximizing accuracy on 10 out of the 14 datasets, and when considering minimizing misclassification costs only, ECSDT performs better on 10 out of the 14 datasets, while when considering both accuracy and misclassification costs, ECSDT was able to obtain higher accuracy on 10 out of the 14 datasets and minimize misclassification costs on 5 out of the 14 datasets. The ECSDT also was able to produce smaller trees when compared with J48, LADTree and ADTree.

Rozhodovácí stromy jsou rozšířenou technikou pro popis dat. Používají se často teké pro predikace. Zajímavým problémemje, že konkrétní distribuce může být popsána jedním či více rozhodovacími stromy.Obvykle nás zajímá co nejjednodušší rozhodovací strom(který budeme nazývat též optimální rozhodovací strom).Tato práce navrhuje rozšíření prořezávácí fáze algoritmů pro rozhodovací stromytak, aby umožňovala více prořezávání. V práci byly zkoumány teoretické i praktické vlastnosti tohoto rozšířeného algoritmu. Jako hlavní teoretický výsledek bylo dokázano, že pro jistou třídu distribucí nalezne algoritmus optimální rozhodovací strom(tj.nejmenší rozhodovací strom, který reprezentuje danou distribuci). V praktických testech bylo zkoumáno, jak je schopen algoritmus rekonstruovat známý strom z dat. Zajímalo nás, zdali dosáhne naše rozšíření zlepšení v počtu správně rekonstruovaných stromů zejména v případě, že data jsou dodatečně velká ( z hlediska počtu záznamů). Tato doměnka byla potvrzena praktickými testy. Obdobný výsledek byl před několika lety prokázán pro Bayesovské sítě. Algoritmus navržený v této disertační práci je polynomiální v počtu listů stromu, který je výstupem hladového algoritmu pro růst stromů, což je vylepšení oproti jednoduchému algoritmu prohledávání všech možných stromů, který je exponenciální.

The Binary Decision Diagram (BDD) method has emerged as an alternative to conventional techniques for performing both qualitative and quantitative analysis of fault trees. BDDs are already proving to be of considerable use in reliability analysis, providing a more efficient means of analysing a system, without the need for the approximations previously used in the traditional approach of Kinetic Tree Theory. In order to implement this technique, a BDD must be constructed from the fault tree, according to some ordering of the fault tree variables. The selected variable ordering has a crucial effect on the resulting BDD size and the number of calculations required for its construction; a bad choice of ordering can lead to excessive calculations and a BDD many orders of magnitude larger than one obtained using an ordering more suited to the tree. Within this thesis a comparison is made of the effectiveness of several ordering schemes, some of which have not previously been investigated. Techniques are then developed for the efficient construction of BDDs from fault trees. The method of Faunet reduction is applied to a set of fault trees and is shown to significantly reduce the size of the resulting BDDs. The technique is then extended to incorporate an additional stage that results in further improvements in BDD size. A fault tree analysis strategy is proposed that increases the likelihood of obtaining a BDD for any given fault tree. This method implements simplification techniques, which are applied to the fault tree to obtain a set of concise and independent subtrees, equivalent to the original fault tree structure. BDDs are constructed for each subtree and the quantitative analysis is developed for the set of BDDs to obtain the top event parameters and the event criticality functions.

Information technology (IT) has become one of the key technologies for economic and social development in any organization. Therefore the management of Information technology incidents, and particularly in the area of resolving the problem very fast, is of concern to Information technology managers. Delays can result when incorrect subjects are assigned to Information technology incident calls: because the person sent to remedy the problem has the wrong expertise or has not brought with them the software or hardware they need to help that user. In the case study used for this work, there are no management checks in place to verify the assigning of incident description subjects. This research aims to develop a method that will tackle the problem of wrongly assigned subjects for incident descriptions. In particular, this study explores the Information technology incident calls database of an oil and gas company as a case study. The approach was to explore the Information technology incident descriptions and their assigned subjects; thereafter the correctly-assigned records were used for training decision tree classification algorithms using Waikato Environment for Knowledge Analysis (WEKA) software. Finally, the records incorrectly assigned a subject by human operators were used for testing. The J48 algorithm gave the best performance and accuracy, and was able to correctly assign subjects to 81% of the records wrongly classified by human operators.

The replication of human intelligence, learning and reasoning by means of computer algorithms is termed Artificial Intelligence (Al) and the interaction of such algorithms with the physical world can be achieved using robotics. The work described in this thesis investigates the applications of concept learning (an approach which takes its inspiration from biological motivations and from survival instincts in particular) to robot control and path planning. The methodology of concept learning has been applied using learning decision trees (DTs) which induce domain knowledge from a finite set of training vectors which in turn describe systematically a physical entity and are used to train a robot to learn new concepts and to adapt its behaviour. To achieve behaviour learning, this work introduces the novel approach of hierarchical learning and knowledge decomposition to the frame of the reactive robot architecture. Following the analogy with survival instincts, the robot is first taught how to survive in very simple and homogeneous environments, namely a world without any disturbances or any kind of "hostility". Once this simple behaviour, named a primitive, has been established, the robot is trained to adapt new knowledge to cope with increasingly complex environments by adding further worlds to its existing knowledge. The repertoire of the robot behaviours in the form of symbolic knowledge is retained in a hierarchy of clustered decision trees (DTs) accommodating a number of primitives. To classify robot perceptions, control rules are synthesised using symbolic knowledge derived from searching the hierarchy of DTs. A second novel concept is introduced, namely that of multi-dimensional fuzzy associative memories (MDFAMs). These are clustered fuzzy decision trees (FDTs) which are trained locally and accommodate specific perceptual knowledge. Fuzzy logic is incorporated to deal with inherent noise in sensory data and to merge conflicting behaviours of the DTs. In this thesis, the feasibility of the developed techniques is illustrated in the robot applications, their benefits and drawbacks are discussed.

The main purpose of this work is to compare classifiers. Random Forest and XGBoost are two popular machine learning algorithms. In this paper, we looked at how they work, compared their features, and obtained accurate results from their robots.

La espectroscopía de emisión óptica es una técnica que permite la identificación de elementos químicos usando el espectro electromagnético que emite un plasma. Con base en la literatura. tiene aplicaciones diversas, por ejemplo: en la identificación de entes estelares, para determinar el punto final de los procesos de plasma en la fabricación de semiconductores o bien, específicamente en este trabajo, se tratan espectros para la determinación de elementos presentes en la degradación de compuestos recalcitrantes. En este documento se identifican automáticamente espectros de elementos tales como He, Ar, N, O, y Hg, en sus niveles de energía uno y dos, mediante técnicas de Machine Learning (ML). En primer lugar, se descargan las líneas de elementos reportadas en el NIST (National Institute of Standards and Technology), después se preprocesan y unifican para los siguientes procesos: a) crear un generador de 84 espectros sintéticos implementado en Python y el módulo ipywidgets de Jupyter Notebook, con las posibilidades de elegir un elemento, nivel de energía, variar la temperatura, anchura a media altura, y normalizar el especto y, b) extraer las líneas para los elementos He, Ar, N, O y Hg en el rango de los 200 nm a 890 nm, posteriormente, se les aplica sobremuestreo para realizar la búsqueda de hiperparámetros para los algoritmos: Decision Tree, Bagging, Random Forest y Extremely Randomized Trees basándose en los principios del diseño de experimentos de aleatorización, replicación, bloqueo y estratificación.

Problembakgrund: Mellan år 1970-2009 utgjordes apoteksmarknaden i Sverige av ett statligt monopol, där Apoteket AB hanterade läkemedelsförsörjning för samtliga landsting i Sverige. År 2009 privatiserades däremot apoteksmarknaden, och landstingen fick nu välja om det skulle hantera läkemedelsförsörjningen i egen regi eller fortsätta upphandla tjänsten till en extern aktör. Åren efter avregleringen har landstingen valt att gå olika vägar, där några valt att fortsätta outsourca denna tjänst och andra har tagit hem tjänsten och hanterar den i egen regi. Med kostnadsbesparingar och vårdkvalitet i fokus för landstingen, är det därför intressant att undersöka varför de hanterar tjänsten olika. Syfte: Syftet med studien är att först kartlägga hur landstingen i Sverige hanterar läkemedelsförsörjningen och därefter undersöka och identifiera vilka kritiska faktorer som finns gällande valet av hanteringssätt. Vidare avser studien att analysera hur valet av hanteringsätt påverkas av dessa kritiska faktorer. Utifrån denna analys är det sedan möjligt att utarbeta en modell för sourcingbeslut gällande läkemedelsförsörjning i svensk hälso- och sjukvård. Metod: I studien genomfördes en surveyundersökning, där avsikten var att utföra strukturerade telefonintervjuer på samtliga landsting i Sverige. Studien utgick ifrån en kvantitativ forskningsstrategi med inslag av kvalitativa delar. Detta för att kartlägga landstingens hanteringssätt av läkemedelsförsörjning, samt undersöka drivkrafter och kritiska faktorer vid valet av hanteringssätt. Slutsats: En beslutsmodell i form av ett beslutsträd utformades för sourcingbeslut gällande läkemedelsförsörjningen för svenska landsting. Beslutsmodellen utgick ifrån tre huvudområden som var kritiska vid valet av hanteringssätt gällande läkemedelsförsörjning – fokus på kärnverksamhet, kostnadsbesparingar och vårdkvalitet. Dessa utgjorde grunden i beslutsmodellen, och var avgörande vid beslutsfattandet gällande hanteringssättet.
Background: During the years of 1970-2009 the pharmacy market In Sweden was run by the government, where Apoteket AB managed drug supply for all counties in Sweden. In 2009, however, the pharmacy market was privatized and the county councils, which are responsible for the Swedish health care, now had to choose whether it would manage the drug supply in-house, or continue to procure the service from an external player. The years after deregulation county councils decided to go different ways with this, where some chose to continue to outsourcing this service and other decided to manage it in-house. With cost savings and quality of care as the main focus of the county councils, it is interesting to examine why they handle this service differently. Purpose: The purpose of the study is to first identify how the county councils in Sweden handle their drug supply, and then examine and identify the critical factors by outsourcing this service or by managing it in-house. Furthermore, the study will analyze how the choice of managing this service in-house or outsource it is affected by these critical factors. Based on this analysis, it is then possible to develop a model for sourcing decisions regarding drug supply in the Swedish health care. Method: The study was conducted using a survey study, where structured telephone interviews were used as a data collection method on the county councils in Sweden. The study was based on a quantitative research strategy, with some qualitative elements. This was considered necessary to map out how the county councils managed their drug supplying, and to examine the driving forces and critical factors in choosing between outsourcing or in-house. Conclusion: A decision model in the form of a decision tree was designed for sourcing decisions regarding drug supply for the Swedish county councils. The decision model was based on three main areas that were established as critical in the selection of management methods regarding the drug supplying – focus on core activities, cost savings and quality of care. These areas formed the basis of the decision model, and were established instrumental in sourcing decisions regarding drug supplying in Swedish health care.

Unscheduled aircraft maintenance causes a lot problems and costs for aircraft operators. This is due to the fact that aircraft cause significant costs if flights have to be delayed or canceled and because spares are not always available at any place and sometimes have to be shipped across the world. Reducing the number of unscheduled maintenance is thus a great costs factor for aircraft operators. This thesis describes three methods for aircraft health monitoring and prediction; one method for system monitoring, one method for forecasting of time series and one method that combines the two other methods for one complete monitoring and prediction process. Together the three methods allow the forecasting of possible failures. The two base methods use decision trees for decision making in the processes and genetic optimization to improve the performance of the decision trees and to reduce the need for human interaction. Decision trees have the advantage that the generated code can be fast and easily processed, they can be altered by human experts without much work and they are readable by humans. The human readability and modification of the results is especially important to include special knowledge and to remove errors, which the automated code generation produced.

Inductive symbolic learning algorithms have been used successfully over the years to build knowledge-based systems. One of these, a decision-tree induction algorithm, has formed the central component in several commercial packages because of its particular efficiency, simplicity, and popularity. However, the decision-tree induction algorithms developed thus far are limited to domains where each decision instance's outcome belongs to only a single decision outcome class. Their goal is merely to specify the properties necessary to distinguish instances pertaining to different decision outcome classes. These algorithms are not readily applicable to many challenging new types of applications in which decision instances have outcomes belonging to more than one decision outcome class (i.e., joint decision outcomes). Furthermore, when applied to domains with a single decision outcome, these algorithms become less efficient as the number of the pre-defined outcome classes increases. The objective of this dissertation is to modify previous decision-tree induction techniques in order to apply them to applications with joint decision outcomes. We propose a new decision-tree induction approach called the Multi-Decision-Tree Induction (MDTI) approach. Data was collected for a patient image retrieval application where more than one prior radiological examination would be retrieved based on characteristics of the current examination and patient status. We present empirical comparisons of the MDTI approach with the Backpropagation network algorithm and the traditional knowledge-engineer-driven knowledge acquisition approach, using the same set of cases. These comparisons are made in terms of recall rate, precision rate, average number of prior examinations suggested, and understandability of the acquired knowledge. The results show that the MDTI approach outperforms the Backpropagation network algorithms and is comparable to the traditional approach in all performance measures considered, while requiring much less learning time than either approach. To gain analytical and empirical insights into MDTI, we have compared this approach with the two best known symbolic learning algorithms (i.e., ID3 and AQ) using data domains with a single decision outcome. It has been found analytically that rules generated by the MDTI approach are more general and supported by more instances in the training set. Four empirical experiments have supported the findings.

A pavement management system (PMS) is a valuable tool and one of the critical elements of the highway transportation infrastructure. Since a vast amount of pavement data is frequently and continuously being collected, updated, and exchanged due to rapidly deteriorating road conditions, increased traffic loads, and shrinking funds, resulting in the rapid accumulation of a large pavement database, knowledge-based expert systems (KBESs) have therefore been developed to solve various transportation problems. This dissertation presents the development of theory and algorithm for a new decision tree induction method, called co-location-based decision tree (CL-DT.) This method will enhance the decision-making abilities of pavement maintenance personnel and their rehabilitation strategies. This idea stems from shortcomings in traditional decision tree induction algorithms, when applied in the pavement treatment strategies. The proposed algorithm utilizes the co-location (co-occurrence) characteristics of spatial attribute data in the pavement database. With the proposed algorithm, one distinct event occurrence can associate with two or multiple attribute values that occur simultaneously in spatial and temporal domains. This research dissertation describes the details of the proposed CL-DT algorithms and steps of realizing the proposed algorithm. First, the dissertation research describes the detailed colocation mining algorithm, including spatial attribute data selection in pavement databases, the determination of candidate co-locations, the determination of table instances of candidate colocations, pruning the non-prevalent co-locations, and induction of co-location rules. In this step, a hybrid constraint, i.e., spatial geometric distance constraint condition and a distinct event-type constraint condition, is developed. The spatial geometric distance constraint condition is a neighborhood relationship-based spatial joins of table instances for many prevalent co-locations with one prevalent co-location; and the distance event-type constraint condition is a Euclidean distance between a set of attributes and its corresponding clusters center of attributes. The dissertation research also developed the spatial feature pruning method using the multi-resolution pruning criterion. The cross-correlation criterion of spatial features is used to remove the nonprevalent co-locations from the candidate prevalent co-location set under a given threshold. The dissertation research focused on the development of the co-location decision tree (CL-DT) algorithm, which includes the non-spatial attribute data selection in the pavement management database, co-location algorithm modeling, node merging criteria, and co-location decision tree induction. In this step, co-location mining rules are used to guide the decision tree generation and induce decision rules. For each step, this dissertation gives detailed flowcharts, such as flowchart of co-location decision tree induction, co-location/co-occurrence decision tree algorithm, algorithm of colocation/co-occurrence decision tree (CL-DT), and outline of steps of SFS (Sequential Feature Selection) algorithm. Finally, this research used a pavement database covering four counties, which are provided by NCDOT (North Carolina Department of Transportation), to verify and test the proposed method. The comparison analyses of different rehabilitation treatments proposed by NCDOT, by the traditional DT induction algorithm and by the proposed new method are conducted. Findings and conclusions include: (1) traditional DT technology can make a consistent decision for road maintenance and rehabilitation strategy under the same road conditions, i.e., less interference from human factors; (2) the traditional DT technology can increase the speed of decision-making because the technology automatically generates a decision-tree and rules if the expert knowledge is given, which saves time and expenses for PMS; (3) integration of the DT and GIS can provide the PMS with the capabilities of graphically displaying treatment decisions, visualizing the attribute and non-attribute data, and linking data and information to the geographical coordinates. However, the traditional DT induction methods are not as quite intelligent as oneâ s expectations. Thus, post-processing and refinement is necessary. Moreover, traditional DT induction methods for pavement M&R strategies only used the non-spatial attribute data. It has been demonstrated from this dissertation research that the spatial data is very useful for the improvement of decision-making processes for pavement treatment strategies. In addition, the decision trees are based on the knowledge acquired from pavement management engineers for strategy selection. Thus, different decision-trees can be built if the requirement changes.
Ph. D.

En la actualidad, las empresas manejan una gran cantidad de información, el cual era inimaginable años atrás, la capacidad de recolectarla es muy impresionante. En consecuencia, para varias empresas esta información se ha convertido en un tema difícil de manejar. Diariamente, las empresas sea del sector, tipo o tamaño que sea, toman decisiones, las cuales la mayoría son decisiones estratégicas que pueden afectar el correcto funcionamiento de la empresa. Es aquí, donde ingresa una de las herramientas más mencionadas en el área de TI: Business Intelligence, este término se refiere al uso de datos en una empresa para facilitar la toma de decisiones, explotar su información, y mejor aún, plantear o predecir escenarios a futuro. El presente trabajo permitirá al área de Marketing de la empresa Iasacorp International, obtener información sobre el comportamiento y hábitos de compra de los clientes, mediante técnicas de minería de datos como Árbol de Decisión y técnicas de análisis predictivo, la cual ayudará a la toma de decisiones para establecer estrategias de venta de las líneas (bisutería, complementos de vestir, accesorios de cabello, etc.) que maneja la empresa y de las próximas compras. De acuerdo a lo planteado anterior mente, la implementación de este tipo de sistemas de información ofrece a la empresa ventajas competitivas, permite a la gerencia analizar y entender mejor la información y por consecuencia tomar mejores decisiones de negocio. At present, companies handle a lot of information, which was unimaginable years ago, the ability to collect it is very impressive. Consequently, for many companies this information has become a difficult issue to handle. Due to the large volume of information we have, instead of being useful you can fall in a failed attempt to give proper use. Every day, companies in any sector, type or size, make decisions, most of which are strategic decisions that may affect the proper functioning of the company. It´s here, where we talk about the most mentioned tools in the area of IT: Business Intelligence, this term refers to the use of data in an enterprise to facilitate decision-making, exploit their information, and better yet, raise or predict scenarios future. This work will allow the area Iasacorp Marketing Company International, information on the behavior and buying habits of customers, through predictive analysis techniques, which will help the decision to establish sales strategies lines (jewelry, clothing, hair accessories, etc.) that manages the company and nearby shopping. According to the points made above, the implementation of such information systems offers companies competitive advantages, allows management to better analyze and understand information and consequently make better business decisions.

Modern biological analysis techniques supply scientists with various forms of data. One category of such data are the so called "expression data". These data indicate the quantities of biochemical compounds present in tissue samples.

Recently, expression data can be generated at a high speed. This leads in turn to amounts of data no longer analysable by classical statistical techniques. Systems biology is the new field that focuses on the modelling of this information.

At present, various methods are used for this purpose. One superordinate class of these meth­ods is machine learning. Methods of this kind had, until recently, predominantly been used for classification and prediction tasks. This neglected a powerful secondary benefit: the ability to induce interpretable models.

Obtaining such models from data has become a key issue within Systems biology. Numerous approaches have been proposed and intensively discussed. This thesis focuses on the examination and exploitation of one basic technique: decision trees.

The concept of comparing sets of decision trees is developed. This method offers the pos­sibility of identifying significant thresholds in continuous or discrete valued attributes through their corresponding set of decision trees. Finding significant thresholds in attributes is a means of identifying states in living organisms. Knowing about states is an invaluable clue to the un­derstanding of dynamic processes in organisms. Applied to metabolite concentration data, the proposed method was able to identify states which were not found with conventional techniques for threshold extraction.

A second approach exploits the structure of sets of decision trees for the discovery of com­binatorial dependencies between attributes. Previous work on this issue has focused either on expensive computational methods or the interpretation of single decision trees ­ a very limited exploitation of the data. This has led to incomplete or unstable results. That is why a new method is developed that uses sets of decision trees to overcome these limitations.

Both the introduced methods are available as software tools. They can be applied consecu­tively or separately. That way they make up a package of analytical tools that usefully supplement existing methods.

By means of these tools, the newly introduced methods were able to confirm existing knowl­edge and to suggest interesting and new relationships between metabolites.

Neuere biologische Analysetechniken liefern Forschern verschiedenste Arten von Daten. Eine Art dieser Daten sind die so genannten "Expressionsdaten". Sie geben die Konzentrationen biochemischer Inhaltsstoffe in Gewebeproben an.

Neuerdings können Expressionsdaten sehr schnell erzeugt werden. Das führt wiederum zu so großen Datenmengen, dass sie nicht mehr mit klassischen statistischen Verfahren analysiert werden können. "System biology" ist eine neue Disziplin, die sich mit der Modellierung solcher Information befasst.

Zur Zeit werden dazu verschiedenste Methoden benutzt. Eine Superklasse dieser Methoden ist das maschinelle Lernen. Dieses wurde bis vor kurzem ausschließlich zum Klassifizieren und zum Vorhersagen genutzt. Dabei wurde eine wichtige zweite Eigenschaft vernachlässigt, nämlich die Möglichkeit zum Erlernen von interpretierbaren Modellen.

Die Erstellung solcher Modelle hat mittlerweile eine Schlüsselrolle in der "Systems biology" erlangt. Es sind bereits zahlreiche Methoden dazu vorgeschlagen und diskutiert worden. Die vorliegende Arbeit befasst sich mit der Untersuchung und Nutzung einer ganz grundlegenden Technik: den Entscheidungsbäumen.

Zunächst wird ein Konzept zum Vergleich von Baummengen entwickelt, welches das Erkennen bedeutsamer Schwellwerte in reellwertigen Daten anhand ihrer zugehörigen Entscheidungswälder ermöglicht. Das Erkennen solcher Schwellwerte dient dem Verständnis von dynamischen Abläufen in lebenden Organismen. Bei der Anwendung dieser Technik auf metabolische Konzentrationsdaten wurden bereits Zustände erkannt, die nicht mit herkömmlichen Techniken entdeckt werden konnten.

Ein zweiter Ansatz befasst sich mit der Auswertung der Struktur von Entscheidungswäldern zur Entdeckung von kombinatorischen Abhängigkeiten zwischen Attributen. Bisherige Arbeiten hierzu befassten sich vornehmlich mit rechenintensiven Verfahren oder mit einzelnen Entscheidungsbäumen, eine sehr eingeschränkte Ausbeutung der Daten. Das führte dann entweder zu unvollständigen oder instabilen Ergebnissen. Darum wird hier eine Methode entwickelt, die Mengen von Entscheidungsbäumen nutzt, um diese Beschränkungen zu überwinden.

Beide vorgestellten Verfahren gibt es als Werkzeuge für den Computer, die entweder hintereinander oder einzeln verwendet werden können. Auf diese Weise stellen sie eine sinnvolle Ergänzung zu vorhandenen Analyswerkzeugen dar.

Mit Hilfe der bereitgestellten Software war es möglich, bekanntes Wissen zu bestätigen und interessante neue Zusammenhänge im Stoffwechsel von Pflanzen aufzuzeigen.

The Polarised Gamma-ray Observer (PoGOLite) is a balloon-borne experiment aimed at measuring the polarisation of hard X-rays from astronomical sources. In the planned flight environment the neutron background is high. A smaller version of PoGOLite, named PoGOLino, was constructed with the goal of measuring the neutron background rates and was launched in March 2013. The signals produced in the detectors of both these instruments give rise to waveforms of different shapes depending on the type of detector the interaction occurred in. A method to distinguish between signal and background waveforms based on their shape has been developed. This was done using a machine learning algorithm called boosted decision trees, implemented in the software package Toolkit for Multivariate Data Analysis (TMVA). By constructing new discriminating variables the classification efficiency was improved. The developed classification will be applied to the measurements taken during the 2013 flight of PoGOLino and the method can also be used for the data analysis of future PoGOLite measurements.

CONSELHO NACIONAL DE DESENVOLVIMENTO CIENTÍFICO E TECNOLÓGICO
A importância da previsão de carga a curto prazo (até uma semana à frente) em crescido recentemente. Com os processos de privatização e implantação de ompetição no setor elétrico brasileiro, a previsão de tarifas de energia vai se tornar extremamente importante. As previsões das cargas elétricas são fundamentais para alimentar as ferramentas analíticas utilizadas na sinalização das tarifas. Em conseqüência destas mudanças estruturais no setor, a variabilidade e a não-estacionaridade das cargas elétricas tendem a aumentar devido à dinâmica dos preços da energia. Em função das mudanças estruturais do setor elétrico, previsores mais autônomos são necessários para o novo cenário que se aproxima. As ferramentas disponíveis no mercado internacional para previsão de carga elétrica requerem uma quantidade significativa de informações on-line, principalmente no que se refere a dados meteorológicos. Como a realidade brasileira ainda não permite o acesso a essas informações será proposto um previsor de carga para o curto-prazo, considerando restrições na aquisição dos dados de temperatura. Logo, tem-se como proposta um modelo de previsão de carga horária de curto prazo (um dia a frente) empregando dados de carga elétrica e dados meteorológicos (temperatura) através de modelos de árvore de decisão. Decidiu-se pelo modelo de árvore de decisão, pois este modelo além de apresentar uma grande facilidade de interpretação dos resultados, apresenta pouquíssima ênfase em sua utilização na área de previsão de carga elétrica.
The importance of load forecasting for the short term (up to one-week ahead) has been steadily growing in the last years. Load forecasts are the basis for the forecasting of energy prices, and the privatisation, and the introduction of competitiveness in the Brazilian electricity sector, have turned price forecasting into an extremely important task. As a consequence of structural changes in the electricity sector, the variability and the non-stationarity of the electrical loads have tended to increase, because of the dynamics of the energy prices. As a consequence of these structural changes, new forecasting methods are needed to meet the new scenarios. The tools that are available for load forecasting in the international market require a large amount of online information, specially information about weather data. Since this information is not yet readily available in Brazil, this thesis proposes a short-term load forecaster that takes into consideration the restrictions in the acquisition of temperature data. A short-term (one-day ahead) forecaster of hourly loads is proposed that combines load data and weather data (temperature), by means of decision tree models. Decision trees were chosen because those models, despite being easy to interpret, have been very rarely used for load forecasting.

Decision trees are one of the most commonly used tools in decision analysis, knowledge representation, machine learning, etc., for its simplicity and interpretability. We consider an extension of dynamic programming approach to process the whole set of decision trees for the given decision table which was previously only attainable by brute-force algorithms. We study decision tables with many-valued decisions (each row may contain multiple decisions) because they are more reasonable models of data in many cases. To address this problem in a broad sense, we consider not only decision trees but also inhibitory trees where terminal nodes are labeled with “̸= decision”. Inhibitory trees can sometimes describe more knowledge from datasets than decision trees. As for cost functions, we consider depth or average depth to minimize time complexity of trees, and the number of nodes or the number of the terminal, or nonterminal nodes to minimize the space complexity of trees. We investigate the multi-stage optimization of trees relative to some cost functions, and also the possibility to describe the whole set of strictly optimal trees. Furthermore, we study the bi-criteria optimization cost vs. cost and cost vs. uncertainty for decision trees, and cost vs. cost and cost vs. completeness for inhibitory trees. The most interesting application of the developed technique is the creation of multi-pruning and restricted multi-pruning approaches which are useful for knowledge representation and prediction. The experimental results show that decision trees constructed by these approaches can often outperform the decision trees constructed by the CART algorithm. Another application includes the comparison of 12 greedy heuristics for single- and bi-criteria optimization (cost vs. cost) of trees. We also study the three approaches (decision tables with many-valued decisions, decision tables with most common decisions, and decision tables with generalized decisions) to handle inconsistency of decision tables. We also analyze the time complexity of decision and inhibitory trees over arbitrary sets of attributes represented by information systems in the frameworks of local (when we can use in trees only attributes from problem description) and global (when we can use in trees arbitrary attributes from the information system) approaches.

碩士
國立臺北護理健康大學
護理助產研究所
99
Purpose: This study explores factors that influence the “rooming in” decision of new mothers. Rooming in is the practice of mothers sharing a hospital room with their newborn infant 24 hours/day postpartum.The author used a cross-sectional binary decision tree to create a classification prediction modelable to predict new mother decision-making. Methodology: This study employed cross-sectional decision tree analysis. Data from a convenience sample of expectant mothers were collected between January and April 2011. The sample included 255 subjects, 180 who delivered vaginally and 75 who delivered via cesarean section. Inclusion criteria included 1) in at least the 37th week of pregnancy at enrollment, 2) single delivery, 3) no abnormalities / comorbidities affecting either infant or mother during pregnancy, labor and postpartum periods, and 4) nocontraindications to mother-child bed sharing, intent to give the infant up for adoption and history of the mother being sexually abused. This study was framed on the theory of planned behavior (TPB). Research instruments used included a self-developed subject demographics questionnaire, a revised fatigue scale, medical environment and staff skills scale, “roomingin” attitudes scale, and “doing the month” attitudes scale. Results:This study used a generalized estimating equation (GEE)and logistic regression to assessfactors influencing new mothers’rooming in decisions. Key findings included: 1) Increased number of postpartum hospitalization days reduced subject rooming in willingness by 42.0% to 60.4% (OR=0.396~ 0.580, p= .002~ <.001); 2) Decreased number of sleeping hours reduced subject rooming in willingness by 23.8% (OR=0.762, p= .005); 3) Each one-point increase in medical environment and staff skills scale score increased subject rooming in willingness by 10.6% (OR= 1.106, p= .005); and 4) Each one-point increase in respect for “doing the month” decreased subject rooming in willingness by 9.7% (OR=0.903, p= .001). All identified factors met significance criteria. Decision tree induction demonstrated the important influence of a mother’s pre-pregnancy preferences overpostpartum rooming in willingness. Subjects who were predisposed to roominginand hada medical environment and staff skills scale score over 30.5 were likely to accept staying with their infant in their room full time. As for subjects who were predisposed against rooming in and in favor of only in-room feeding or daytime contact, those against rooming in altogether or for in-room feeding onlyand who had a medical environment and staff skills scale score over 18.5 all refused the 24-hour room-in option. Subjects predisposed to in-room daytime contact onlywith a rooming in attitudes scale score over 30.5, a “doing the month” attitudes scale score less than 27.5 and a medical environment and staff skills scale score over 27.5 ultimately accepted the rooming in option. The predictive power of the decision tree attained 87.9%. Conclusions / Implications for Practice: Factors that influence new mother attitudes toward staying full-time with their infant in the hospital room are influenced strongly by personal attitudes and preferences. These, in turn, reflectsuch subjective influences as perceived medical environment quality, medical staff skill, adequate sleep, respect for traditional “doing the month” mores and other external variables that affect length of postpartum hospitalization. Decision tree analysis found mothers’ predisposition toward or against rooming in to be the most important factor influencing their ultimate rooming in decision. Thus, predisposition was the behavior motivation in the research structure. Secondary factors of influence identified in this study included mothers’ subjective opinions regarding medical environment quality and medical staff skill, positive / negative feelings toward rooming in, and level of intent to follow “doing the month” mores. This study demonstrated the influence of these factors on new mother attitudes toward and willingness to share a hospital room with their infant 24 hours/day postpartum. We recommend further stressing the importance of postpartum rooming in as well as presenting important supplementary information (particularly with regard to negative rooming in attitudes) in prenatal education and clinical nursing instruction in order to increase rooming in willingness and rates among postpartum mothers. Findings may also be incorporated into cultural care practices related to “doing the month” and adapted to the needs of mother-centric nursing care in order to help new mothers fulfill their care role. Finally we recommend that pediatricians coordinate formally with maternity ward staff to provide a more consistent and comprehensive message to expecting mothers regarding the importance and benefits of roomingin. Key Words: rooming in, binary decision tree, doing the month, rooming in attitudes, nurse support of rooming in

碩士
國立清華大學
原子科學系
89
Abstract Our purpose of this research is to utilize the measures of orthogonal projection and tree decisions on vehicle license plate recognition (LPR). Our research uses orthogonal projection as the core of character recognition. In order to raise the recognition rate, tree decisions based on features of character contours and projection profiles are employed. This thesis consists of three main parts. The first part is to locate the license plate in an image. The second part is to detect and segment each character of the license plate. The third part is on character recognition of license plates. Character recognition is more concerned in our research. Horizontal and vertical projections on orthogonal axes are adopted as comparisons between input characters and standard ones. An argument called cumulative difference values (CDV) is introduced to give a solution from standard database. Tree decisions are added to assist in distinguishing characters. The recognition rate from 59 images is 86.44%, and the recognition rate of 312 characters is 99.68%. Keywords: license plate recognition, LPR, optical character recognition (OCR), character segmentation, orthogonal projection, cumulative difference values (CDV), tree decisions.

碩士
國立成功大學
電腦與通信工程研究所
100
High Efficiency Video Coding (HEVC) is a ongoing video coding standard which is currently under the joint development of ISO/IEC MPEG and ITU-T VCEG. HEVC is also known as the successor of H.264 video coding standard and is expected to be a popular next-generation video codec in the future. HEVC replaces macroblock scheme with coding units (CUs) in the form of quad-tree structure. The encoding process examines all possible CUs recursively on the quad-tree. This process can evaluate the coding performance for the variable sized CUs and retain the best partitions of the CUs. HEVC can provide higher compression ratio compared to H.264/AVC standard; however, the coding complexity is dramatically increased as well. In this thesis, a fast algorithm for coding unit decision is proposed to reduce the burden of the encoding time in HEVC. The proposed algorithm exploits the temporal correlation in the neighboring frames of a video sequence to avoid the unnecessary examinations on CU quad-trees. In addition, based on an adaptive threshold, the best prediction mode is early determined to SKIP mode for reducing the exhaustive evaluations at prediction stage. The performance of the proposed algorithm is veri ed through the test model for HEVC, HM 5.0. The experimental results show that the proposed algorithm can averagely achieve about 27%, 33%, 20%, and 21% total time encoding time reduction under Low-Delay High Effi ciency, Low-Delay Low Complexity, Random-Access High Effi ciency, and Random-Access Low Complexity con gurations respectively with a negligible degradation of coding performance.

碩士
南華大學
資訊管理學系碩士班
90
In this thesis, we propose an image clustering method based on CLTree for image segmentation. CLTree is a clustering algorithm that uses decision-tree technique. It’s quit different from existing clustering methods, and it finds clusters without making any prior assumptions or any input parameters. Whether a clustering is good or bad depends on the user's subjective judgment, so we offer three image segmentation results. The experimental results reveal that all of them perform well.

博士
國立中央大學
資訊管理研究所
98
Classification is one of the most important research domains in data mining. Among the existing classifiers, decision trees are probably the most popular and commonly-used classification models. Most of the decision tree algorithms aimed to maximize the classification accuracy and minimize the classification error. However, in many real-world applications, there are various types of cost or resource consumption involved in both the induction of decision tree and the classification of future instance. Furthermore, the problem we face may require us to complete a classification task with limited resource. Therefore, how to build an optimum decision tree with resource constraint becomes an important issue. In this study, we first propose two algorithms which are improved versions of traditional TDIDT(Top-Down Induction on Decision Trees) algorithms. Then, we adopt a brand new approach to deal with multiple resource constraints. This approach extracts association classification rules from training dataset first, and then builds a decision tree from the extracted rules. Empirical evaluations were carried out using real datasets, and the results indicated that the proposed methods can achieve satisfactory results in handling data under different resource constraints.

碩士
國立嘉義大學
企業管理學系
106
In recent years, financial literacy of citizens has been improving. Furthermore, financial investment channels have likewise multiplied. Most investment tools all need a lot of financial know-how in order to obtain steady profits. Compared to other financial tools, mutual fund risks and barriers to entry are relatively low. The total number of kinds of mutual funds have been increasing yearly and within the many mutual funds available, picking the right fund and strategy to take as the best investment methods are what investors focus on. Every mutual fund has a set of efficiency benchmark. This study analyzes and discusses at the local mutual fund market and uses efficiency benchmark data from 2012 to 2017 of Taiwan’s local stock type and global investment stock type mutual funds. The research uses data mining to analyze the data from these benchmarks and looks for selection and manipulation strategies that can be applied to the mutual funds. Through data mining decision tree analysis, the study categorizes the mutual funds into three types: buy, hold, and sell. This research uses maximum return to explore the problem of investment strategy on mutual funds. Data analysis results help most investors to understand mutual fund strategy and the meaning of each index in order to minimize losses in the mutual fund market.

碩士
國立虎尾科技大學
資訊工程研究所
102
High Efficiency Video Coding (HEVC) is recently developed for ultra high definition video compression technique, which provides a higher compression ratio and throughput compared with previously video compression standard H.264/AVC. Therefore, this technique is widely used to limited bandwidth network transmission and confined storage space. In order to obtain the higher compression ratio and maintain video quality, which provides variable block partition and mode prediction for HEVC encoder. If each block is computed during the mode decision process, a lot of encoding time is consumed. It makes limiting the applicability in real time for HEVC. Hence, there are many fast algorithms proposed to eliminate the block partition or mode prediction. In natural videos, the neighbor blocks have high correlation with current block, by which the reference block method is studied to terminate or eliminate the block or mode prediction. This method uses the lower computation of mode reduction to obtain a best compression ratio and time saving. Therefore, that is widely proposed for HEVC fast algorithm. On the other hand, the non-reference method has been proposed by extracting the feature of video frames. But the non-reference method predict the terminated condition. This thesis, proposes two quad-tree depth decision methods : one is the reference method and the other one non-reference method for depth-correlation and edge strength detection method, respectively. In reference block method, we find the correlation of up to 90% correlation with the co-located coding tree block (CTB) in the previous frame. Therefore, we use the co-located CTB depth information to limit the depth partition of CTB. Different from the previously proposed method, the proposed method adopts the extension of partition depth by one level. But it is poor prediction in fast moving object sequence or change scene. The fast moving and changing scenes are lower correlation between frames. Based on aforementioned disadvantage, the edge strength detection method is proposed to detect the structure variation of CTB to predict the encoded depth. Since this method does not require the reference to neighbor block, a better prediction with variation video sequence can be obtained. But it makes the poor prediction for unobvious edge video. For example, in dark videos, the edge are not obvious and the proposed algorithm makes the poor prediction of depth level. Finally, the proposed fast methods are implemented in HM 10.1 model to demonstrate the efficiency of our algorithm. The proposed edge density detection method can obtain 23.1% of time savings with BD-bitrate close to 0.28% on average and depth-correlation method can provide about 21.1% of time savings and BD-bitrate increase of 0.17% on average.