Siga este link para ver outros tipos de publicações sobre o tema: Classification of biomedical time series.
Teses / dissertações sobre o tema "Classification of biomedical time series"
Crie uma referência precisa em APA, MLA, Chicago, Harvard, e outros estilos
Veja os 50 melhores trabalhos (teses / dissertações) para estudos sobre o assunto "Classification of biomedical time series".
Ao lado de cada fonte na lista de referências, há um botão "Adicionar à bibliografia". Clique e geraremos automaticamente a citação bibliográfica do trabalho escolhido no estilo de citação de que você precisa: APA, MLA, Harvard, Chicago, Vancouver, etc.
Você também pode baixar o texto completo da publicação científica em formato .pdf e ler o resumo do trabalho online se estiver presente nos metadados.
Veja as teses / dissertações das mais diversas áreas científicas e compile uma bibliografia correta.
1
Rajan, Jebu Jacob. "Time series classification". Thesis, University of Cambridge, 1994. http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.339538.
Texto completo da fonte
2
Matam, Basava R. "Watermarking biomedical time series data". Thesis, Aston University, 2009. http://publications.aston.ac.uk/15351/.
Texto completo da fonteResumo:
This thesis addresses the problem of information hiding in low dimensional digital data focussing on issues of privacy and security in Electronic Patient Health Records (EPHRs). The thesis proposes a new security protocol based on data hiding techniques for EPHRs. This thesis contends that embedding of sensitive patient information inside the EPHR is the most appropriate solution currently available to resolve the issues of security in EPHRs. Watermarking techniques are applied to one-dimensional time series data such as the electroencephalogram (EEG) to show that they add a level of confidence (in terms of privacy and security) in an individual’s diverse bio-profile (the digital fingerprint of an individual’s medical history), ensure belief that the data being analysed does indeed belong to the correct person, and also that it is not being accessed by unauthorised personnel. Embedding information inside single channel biomedical time series data is more difficult than the standard application for images due to the reduced redundancy. A data hiding approach which has an in built capability to protect against illegal data snooping is developed. The capability of this secure method is enhanced by embedding not just a single message but multiple messages into an example one-dimensional EEG signal. Embedding multiple messages of similar characteristics, for example identities of clinicians accessing the medical record helps in creating a log of access while embedding multiple messages of dissimilar characteristics into an EPHR enhances confidence in the use of the EPHR. The novel method of embedding multiple messages of both similar and dissimilar characteristics into a single channel EEG demonstrated in this thesis shows how this embedding of data boosts the implementation and use of the EPHR securely.
3
Khessiba, Souhir. "Stratégies d’optimisation des hyper-paramètres de réseaux de neurones appliqués aux signaux temporels biomédicaux". Electronic Thesis or Diss., Institut polytechnique de Paris, 2024. http://www.theses.fr/2024IPPAE003.
Texto completo da fonteResumo:
Cette thèse est axée sur l'optimisation des hyperparamètres des réseaux de neurones à convolution (CNN) dans le domaine médical, proposant une approche innovante visant à améliorer la performance des modèles décisionnels dans le domaine biomédical. Grâce à l'utilisation d'une approche hybride, GS-TPE, pour ajuster efficacement les hyperparamètres des modèles de réseaux de neurones complexes , cette recherche a démontré des améliorations significatives dans la classification des signaux biomédicaux temporels, à savoir les états de vigilance, à partir de signaux physiologiques tels que l'électroencéphalogramme (EEG). De plus, grâce à l'introduction d'une nouvelle architecture de DNN, STGCN, pour la classification de gestes associés à des pathologies telles que l'arthrose du genou et la maladie de Parkinson à partir d'analyses vidéo de la marche, ces travaux offrent de nouvelles perspectives pour l'amélioration du diagnostic et de la prise en charge médicale grâce aux progrès dans le domaine de l'IAThis thesis focuses on optimizing the hyperparameters of convolutional neural networks (CNNs) in the medical domain, proposing an innovative approach to improve the performance of decision-making models in the biomedical field. Through the use of a hybrid approach, GS-TPE, to effectively adjust the hyperparameters of complex neural network models, this research has demonstrated significant improvements in the classification of temporal biomedical signals, such as vigilance states, from physiological signals such as electroencephalogram (EEG). Furthermore, by introducing a new DNN architecture, STGCN, for the classification of gestures associated with pathologies such as knee osteoarthritis and Parkinson's disease from video gait analysis, these works offer new perspectives for enhancing medical diagnosis and management through advancements in artificial intelligence
4
Al-Wasel, Ibrahim A. "Spectral analysis for replicated biomedical time series". Thesis, Lancaster University, 1993. http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.412585.
Texto completo da fonte
5
Esling, Philippe. "Multiobjective time series matching and classification". Paris 6, 2012. http://www.theses.fr/2012PA066704.
Texto completo da fonteResumo:
Plusieurs millions d’années d’évolution génétique ont façonné notre système auditif, permettant d’effectuer une discrimination flexible des événements acoustiques. Nous pouvons ainsi traiter simultanément plusieurs échelles de perception contradictoires de manière multidimensionnelle. De plus, nous avons une capacité à extraire une structure cohérente à partir de formes temporelles. Nous montrons qu’en émulant ces mécanismes dans nos choix algorithmiques, nous pouvons créer des approches efficaces de recherche et classification, dépassant le cadre des problématiques musicales. Nous introduisons ainsi le problème de MultiObjective Time Series (MOTS) et proposons un algorithme efficace pour le résoudre. Nous introduisons deux paradigmes innovants de recherche sur les fichiers audio. Nous introduisons un nouveau paradigme de classification basé sur la domination d'hypervolume, appelé HyperVolume-MOTS (HV-MOTS). Ce système étudie le comportement de la classe entière par sa distribution et sa diffusion sur l’espace d’optimisation. Nous montrons une amélioration sur les méthodes de l’état de l’art sur un large éventail de problèmes scientifiques. Nous présentons ainsi un système d’identification biométrique basée sur les sons produit par les battements de coeur, atteignant des taux d’erreur équivalents à d’autres caractéristiques biométriques. Ces résultats sont confirmés par le l'ensemble de données cardiaques de l’étude d’isolation Mars500. Enfin, nous étudions le problème de la génération de mélanges sonores orchestraux imitant au mieux une cible audio donnée. L'algorithme de recherche basé sur le problème MOTS nous permet d’obtenir un ensemble de solutions efficacesMillions of years of genetic evolution have shaped our auditory system, allowing to discriminate acoustic events in a flexible manner. We can perceptually process multiple de-correlated scales in a multidimensional way. In addition, humans have a natural ability to extract a coherent structure from temporal shapes. We show that emulating these mechanisms in our algorithmic choices, allow to create efficient approaches to perform matching and classification, with a scope beyond musical issues. We introduce the problem of multiobjective Time Series (MOTS) and propose an efficient algorithm to solve it. We introduce two innovative querying paradigms on audio files. We introduce a new classification paradigm based on the hypervolume dominated by different classes called hypervolume-MOTS (HV-MOTS). This system studies the behavior of the whole class by its distribution and spread over the optimization space. We show an improvement over the state of the art methods on a wide range of scientific problems. We present a biometric identification systems based on the sounds produced by heartbeats. This system is able to reach low error rates equivalent to other biometric features. These results are confirmed by the extensive cardiac data set of the Mars500 isolation study. Finally, we study the problem of generating orchestral mixtures that could best imitate a sound target. The search algorithm based on MOTS problem allows to obtain a set of solutions to approximate any audio source
6
Nunes, Neuza Filipa Martins. "Algorithms for time series clustering applied to biomedical signals". Master's thesis, Faculdade de Ciências e Tecnologia, 2011. http://hdl.handle.net/10362/5666.
Texto completo da fonteResumo:
Thesis submitted in the fulfillment of the requirements for the Degree of Master in Biomedical EngineeringThe increasing number of biomedical systems and applications for human body understanding creates a need for information extraction tools to use in biosignals. It’s important to comprehend the changes in the biosignal’s morphology over time, as they often contain critical information on the condition of the subject or the status of the experiment. The creation of tools that automatically analyze and extract relevant attributes from biosignals, providing important information to the user, has a significant value in the biosignal’s processing field. The present dissertation introduces new algorithms for time series clustering, where we are able to separate and organize unlabeled data into different groups whose signals are similar to each other. Signal processing algorithms were developed for the detection of a meanwave, which represents the signal’s morphology and behavior. The algorithm designed computes the meanwave by separating and averaging all cycles of a cyclic continuous signal. To increase the quality of information given by the meanwave, a set of wave-alignment techniques was also developed and its relevance was evaluated in a real database. To evaluate our algorithm’s applicability in time series clustering, a distance metric created with the information of the automatic meanwave was designed and its measurements were given as input to a K-Means clustering algorithm. With that purpose, we collected a series of data with two different modes in it. The produced algorithm successfully separates two modes in the collected data with 99.3% of efficiency. The results of this clustering procedure were compared to a mechanism widely used in this area, which models the data and uses the distance between its cepstral coefficients to measure the similarity between the time series.The algorithms were also validated in different study projects. These projects show the variety of contexts in which our algorithms have high applicability and are suitable answers to overcome the problems of exhaustive signal analysis and expert intervention. The algorithms produced are signal-independent, and therefore can be applied to any type of signal providing it is a cyclic signal. The fact that this approach doesn’t require any prior information and the preliminary good performance make these algorithms powerful tools for biosignals analysis and classification.
7
Lines, Jason. "Time Series classification through transformation and ensembles". Thesis, University of East Anglia, 2015. https://ueaeprints.uea.ac.uk/53360/.
Texto completo da fonteResumo:
The problem of time series classification (TSC), where we consider any real-valued ordered data a time series, offers a specific challenge. Unlike traditional classification problems, the ordering of attributes is often crucial for identifying discriminatory features between classes. TSC problems arise across a diverse range of domains, and this variety has meant that no single approach outperforms all others. The general consensus is that the benchmark for TSC is nearest neighbour (NN) classifiers using Euclidean distance or Dynamic Time Warping (DTW). Though conceptually simple, many have reported that NN classifiers are very difficult to beat and new work is often compared to NN classifiers. The majority of approaches have focused on classification in the time domain, typically proposing alternative elastic similarity measures for NN classification. Other work has investigated more specialised approaches, such as building support vector machines on variable intervals and creating tree-based ensembles with summary measures. We wish to answer a specific research question: given a new TSC problem without any prior, specialised knowledge, what is the best way to approach the problem? Our thesis is that the best methodology is to first transform data into alternative representations where discriminatory features are more easily detected, and then build ensemble classifiers on each representation. In support of our thesis, we propose an elastic ensemble classifier that we believe is the first ever to significantly outperform DTW on the widely used UCR datasets. Next, we propose the shapelet-transform, a new data transformation that allows complex classifiers to be coupled with shapelets, which outperforms the original algorithm and is competitive with DTW. Finally, we combine these two works with with heterogeneous ensembles built on autocorrelation and spectral-transformed data to propose a collective of transformation-based ensembles (COTE). The results of COTE are, we believe, the best ever published on the UCR datasets.
8
Ghalwash, Mohamed. "Interpretable Early Classification of Multivariate Time Series". Diss., Temple University Libraries, 2013. http://cdm16002.contentdm.oclc.org/cdm/ref/collection/p245801coll10/id/239730.
Texto completo da fonteResumo:
Computer and Information SciencePh.D.
Recent advances in technology have led to an explosion in data collection over time rather than in a single snapshot. For example, microarray technology allows us to measure gene expression levels in different conditions over time. Such temporal data grants the opportunity for data miners to develop algorithms to address domain-related problems, e.g. a time series of several different classes can be created, by observing various patient attributes over time and the task is to classify unseen patient based on his temporal observations. In time-sensitive applications such as medical applications, some certain aspects have to be considered besides providing accurate classification. The first aspect is providing early classification. Accurate and timely diagnosis is essential for allowing physicians to design appropriate therapeutic strategies at early stages of diseases, when therapies are usually the most effective and the least costly. We propose a probabilistic hybrid method that allows for early, accurate, and patient-specific classification of multivariate time series that, by training on a full time series, offer classification at a very early time point during the diagnosis phase, while staying competitive in terms of accuracy with other models that use full time series both in training and testing. The method has attained very promising results and outperformed the baseline models on a dataset of response to drug therapy in Multiple Sclerosis patients and on a sepsis therapy dataset. Although attaining accurate classification is the primary goal of data mining task, in medical applications it is important to attain decisions that are not only accurate and obtained early, but can also be easily interpreted which is the second aspect of medical applications. Physicians tend to prefer interpretable methods rather than black-box methods. For that purpose, we propose interpretable methods for early classification by extracting interpretable patterns from the raw time series to help physicians in providing early diagnosis and to gain insights into and be convinced about the classification results. The proposed methods have been shown to be more accurate and provided classifications earlier than three alternative state-of-the-art methods when evaluated on human viral infection datasets and a larger myocardial infarction dataset. The third aspect has to be considered for medical applications is the need for predictions to be accompanied by a measure which allows physicians to judge about the uncertainty or belief in the prediction. Knowing the uncertainty associated with a given prediction is especially important in clinical diagnosis where data mining methods assist clinical experts in making decisions and optimizing therapy. We propose an effective method to provide uncertainty estimate for the proposed interpretable early classification methods. The method was evaluated on four challenging medical applications by characterizing decrease in uncertainty of prediction. We showed that our proposed method meets the requirements of uncertainty estimates (the proposed uncertainty measure takes values in the range [0,1] and propagates over time). To the best of our knowledge, this PhD thesis will have a great impact on the link between data mining community and medical domain experts and would give physicians sufficient confidence to put the proposed methods into real practice.
Temple University--Theses
9
Owsley, Lane M. D. "Classification of transient events in time series /". Thesis, Connect to this title online; UW restricted, 1998. http://hdl.handle.net/1773/5989.
Texto completo da fonte
10
Botsch, Michael-Felix. "Machine learning techniques for time series classification". Göttingen Cuvillier, 2009. http://d-nb.info/994721455/04.
Texto completo da fonte
11
Hu, Wei Long. "Candlestick pattern classification in financial time series". Thesis, University of Macau, 2018. http://umaclib3.umac.mo/record=b3950658.
Texto completo da fonte
12
Zhang, Fuli. "Spectral classification of high-dimensional time series". Diss., University of Iowa, 2018. https://ir.uiowa.edu/etd/6531.
Texto completo da fonteResumo:
In this era of big data, multivariate time-series (MTS) data are prevalent in diverse domains and often high dimensional. However, there have been limited studies of building a capable classifier with MTS via classical machine learning methods that can deal with the double curse of dimensionality due to high variable dimension and long time series (large sample size). In this thesis, we propose two approaches to address this problem for multiclass classification with high dimensional MTS.
Both approaches leverage the dynamics of an MTS captured by non-parametric modeling of its spectral density function. In the first approach, we introduce the reduced-rank spectral classifier (RRSC), which utilizes low-rank estimation and some new discrimination functions. We illustrate the efficacy of the RRSC with both simulations and real applications. For binary classification, we establish the consistency of the RRSC and provide an asymptotic formula for the misclassification error rates, under some regularity conditions. The second approach concerns the development of the random projection ensemble classifier for time series (RPECTS). This method first applies dimension reduction in the time domain via projecting the time-series variables into some low dimensional space, followed by measuring the disparity via some novel base classifier between the data and the candidate generating processes in the projected space.
We assess the classification performance of our new approaches by simulations and compare them with some existing methods using real applications. Finally, we elaborate two R packages that implement the aforementioned methods.
13
Dachraoui, Asma. "Cost-Sensitive Early classification of Time Series". Thesis, Université Paris-Saclay (ComUE), 2017. http://www.theses.fr/2017SACLA002/document.
Texto completo da fonteResumo:
Dans de nombreux domaines dans lesquels les mesures ou les données sont disponibles séquentiellement, il est important de savoir décider le plus tôt possible, même si c’est à partir d’informations encore incomplètes. C’est le cas par exemple en milieu hospitalier où l’apprentissage de règles de décision peut se faire à partir de cas complètement documentés, mais où, devant un nouveau patient, il peut être crucial de prendre une dé- cision très rapidement. Dans ce type de contextes, un compromis doit être optimisé entre la possibilité d’arriver à une meilleure décision en attendant des mesures supplé- mentaires, et le coût croissant associé à chaque nouvelle mesure. Nous considérons dans cette thèse un nouveau cadre général de classification précoce de séries temporelles où le coût d’attente avant de prendre une décision est explicitement pris en compte lors de l’optimisation du compromis entre la qualité et la précocité de prédictions. Nous proposons donc un critère formel qui exprime ce compromis, ainsi que deux approches différentes pour le résoudre. Ces approches sont intéressantes et apportent deux propriétés désirables pour décider en ligne : (i) elles estiment en ligne l’instant optimal dans le futur où une minimisation du critère peut être prévue. Elles vont donc au-delà des approches classiques qui décident d’une façon myope, à chaque instant, d’émettre une prédiction ou d’attendre plus d’information, (ii) ces approches sont adaptatives car elles prennent en compte les propriétés de la série temporelle en entrée pour estimer l’instant optimal pour la classifier. Des expériences extensives sur des données contrôlées et sur des données réelles montrent l’intérêt de ces approches pour fournir des prédictions précoces, fiables, adaptatives et non myopes, ce qui est indispensable dans de nombreuses applicationsEarly classification of time series is becoming increasingly a valuable task for assisting in decision making process in many application domains. In this setting, information can be gained by waiting for more evidences to arrive, thus helping to make better decisions that incur lower misclassification costs, but, meanwhile, the cost associated with delaying the decision generally increases, rendering the decision less attractive. Making early predictions provided that are accurate requires then to solve an optimization problem combining two types of competing costs. This thesis introduces a new general framework for time series early classification problem. Unlike classical approaches that implicitly assume that misclassification errors are cost equally and the cost of delaying the decision is constant over time, we cast the the problem as a costsensitive online decision making problem when delaying the decision is costly. We then propose a new formal criterion, along with two approaches that estimate the optimal decision time for a new incoming yet incomplete time series. In particular, they capture the evolutions of typical complete time series in the training set thanks to a segmentation technique that forms meaningful groups, and leverage these complete information to estimate the costs for all future time steps where data points still missing. These approaches are interesting in two ways: (i) they estimate, online, the earliest time in the future where a minimization of the criterion can be expected. They thus go beyond the classical approaches that myopically decide at each time step whether to make a decision or to postpone the call one more time step, and (ii) they are adaptive, in that the properties of the incoming time series are taken into account to decide when is the optimal time to output a prediction. Results of extensive experiments on synthetic and real data sets show that both approaches successfully meet the behaviors expected from early classification systems
14
Omer, Mohamoud. "Estimation of regularity and synchronism in parallel biomedical time series". Phd thesis, Univerzitet u Novom Sadu, Fakultet tehničkih nauka u Novom Sadu, 2017. https://www.cris.uns.ac.rs/record.jsf?recordId=101879&source=NDLTD&language=en.
Texto completo da fonteResumo:
Objectives: Self-monitoring in health applications has already been recognized as a part of the mobile crowdsensing concept, where subjects, equipped with adequate sensors, share and extract information for personal or common benefit. Limited data transmission resources force a local analysis at wearable devices, but it is incompatible with analytical tools that require stationary and artifact-free data. The key objective of this thesis is to explain a computationally efficient binarized cross-approximate entropy, (X)BinEn, for blind cardiovascular signal processing in environments where energy and processor resources are limited.Methods: The proposed method is a descendant of cross-approximate entropy ((X)ApEn). It operates over binary differentially encoded data series, split into m-sized binary vectors. Hamming distance is used as a distance measure, while a search for similarities is performed over the vector sets, instead of over the individual vectors. The procedure is tested in laboratory rats exposed to shaker and restraint stress and compared to the existing (X)ApEn results.Results: The number of processor operations is reduced. (X)BinEn captures entropy changes similarly to (X)ApEn. The coding coarseness has an adverse effect of reduced sensitivity, but it attenuates parameter inconsistency and binary bias. A special case of (X)BinEn is equivalent to Shannon entropy. A binary conditional m=1 entropy is embedded into the procedure and can serve as a complementary dynamic measure.Conclusion: (X)BinEn can be applied to a single time series as auto-entropy or, more generally, to a pair of time series, as cross-entropy. It is intended for mobile, battery operated self-attached sensing devices with limited power and processor resources.Cilj: Snimanje sopstvenih zdravstveih prametara je postalo deo koncepta mobilnog ‘crowdsensing-a’ prema kojem učesnici sa nakačenim senzorima skupljaju i dele informacije, na ličnu ili opštu dobrobit. Međutim, ograničenja u prenosu podataka dovela su do koncepta lokalne obrade (na licu mesta). To je pak nespojivo sa uobičajenim metodama za koje je potrebno da podaci koji se obrađuju budu stacionarni i bez artefakata. Ključni deo ove teze je opis procesorski nezahtevne binarizovane unakrsne aproksimativne entropije (X)BinEn koja omogućava analizu kardiovaskularnih podataka bez prethodne predobrade, u uslovima ograničenog napajanja i procesorskih resursa.Metoda: (X)BinEn je nastao razradom postojećeg postupka unakrsne entropije ((X)ApEn). Definisan je nad binarnim diferencijalno kodovanim vremenskim nizovima, razdeljenim u binarne vektore dužine m. Za procenu razmaka između vektora koristi se Hemingovo rastojanje, a sličnost vektora se ne procenjuje između svakog vektora pojedinačno, već između skupova vektora. Procedura je testirana nad laboratorijskim pacovima izloženim različitim vrstova stresova i upoređena sa postojećim rezultatima.Rezultati: Broj potrebnih procesorskih operacija je značajno smanjen. (X)BinEn registruje promene entropije slično (X)ApEn. Beskonačno klipovanje je gruba kvantizacija i za posledicu ima smanjenu osetljivost na promene, ali, sa druge strane, prigušuje binarnu asimetriju i nekonzistentnan uticaj parametara. Za određeni skup parametara (X)BinEn je ekvivalentna Šenonovoj entropiji. Uslovna binarna m=1 entropija automatski se dobija kao uzgredni product binarizovane entropije, i može da se iskoristi kao komplementarna dinamička mera.Zaključak: (X)BinEn može da se koristi za jedan vremenski niz, kao auto-entropija, ili, u opštem slučaju, za dva vremenska niza kao unakrsna entropija. Namenjena je mobilnim uređajima sa baterijskim napajanjem za individualne korisnike, to jest za korisnike sa ograničenim napajanjem i procesorskim resursima.
15
Santos, Rui Pedro Silvestre dos. "Time series morphological analysis applied to biomedical signals events detection". Master's thesis, Faculdade de Ciências e Tecnologia, 2011. http://hdl.handle.net/10362/10227.
Texto completo da fonteResumo:
Dissertation submitted in the fufillment of the requirements for the Degree of Master in Biomedical EngineeringAutomated techniques for biosignal data acquisition and analysis have become increasingly powerful, particularly at the Biomedical Engineering research field. Nevertheless, it is verified the need to improve tools for signal pattern recognition and classification systems, in which the detection of specific events and the automatic signal segmentation are preliminary processing steps. The present dissertation introduces a signal-independent algorithm, which detects significant events in a biosignal. From a time series morphological analysis, the algorithm computes the instants when the most significant standard deviation discontinuities occur, segmenting the signal. An iterative optimization step is then applied. This assures that a minimal error is achieved when modeling these segments with polynomial regressions. The adjustment of a scale factor gives different detail levels of events detection. An accurate and objective algorithm performance evaluation procedure was designed. When applied on a set of synthetic signals, with known and quantitatively predefined events, an overall mean error of 20 samples between the detected and the actual events showed the high accuracy of the proposed algorithm. Its ability to perform the detection of signal activation onsets and transient waveshapes was also assessed, resulting in higher reliability than signal-specific standard methods. Some case studies, with signal processing requirements for which the developed algorithm can be suitably applied, were approached. The algorithm implementation in real-time, as part of an application developed during this research work, is also reported. The proposed algorithm detects significant signal events with accuracy and significant noise immunity. Its versatile design allows the application in different signals without previous knowledge on their statistical properties or specific preprocessing steps. It also brings added objectivity when compared with the exhaustive and time-consuming examiner analysis. The tool introduced in this dissertation represents a relevant contribution in events detection, a particularly important issue within the wide digital biosignal processing research field.
16
Gordon, Kerry. "Modelling and monitoring of medical time series". Thesis, University of Nottingham, 1986. http://eprints.nottingham.ac.uk/12369/.
Texto completo da fonteResumo:
In this thesis we examine several extensions to the dynamic linear model framework, outlined by Harrison and Stevens (1976), in order to adapt these models for use in the on-line analysis of medical time series that arise from routine clinical settings. The situation with which we are most concerned is that where we are monitoring individual patients and wish to detect abrupt changes in the patient's condition as soon as possible. A detailed background to the study and application of dynamic linear models is given, and other techniques for time series monitoring are also discussed when appropriate. We present a selection of specific models that we feel may prove to be of practical use in the modelling and monitoring of medical time series, and we illustrate how these models may be utilized in order to distinguish between a variety of alternative changepoint-types. The sensitivity of these models to the specification of prior information is examined in detail. The medical background to the time series examined requires the development of models and techniques enabling us to analyze generally unequally-spaced time series. We test the performance of the resulting models and techniques using simulated data. We then attempt to build a framework for bivariate time series modelling, allowing, once more, for the possibility of unequally spaced data. In particular, we suggest mechanisms whereby causality and feedback may be introduced into such models. Finally, we report on several applications of this methodology to actual medical time series arising in various contexts including kidney and bone-marrow transplantation and foetal heart monitoring.
17
Bostrom, Aaron. "Shapelet transforms for univariate and multivariate time series classification". Thesis, University of East Anglia, 2018. https://ueaeprints.uea.ac.uk/67270/.
Texto completo da fonteResumo:
Time Series Classification (TSC) is a growing field of machine learning research. One particular algorithm from the TSC literature is the Shapelet Transform (ST). Shapelets are a phase independent subsequences that are extracted from times series to form discriminatory features. It has been shown that using the shapelets to transform the datasets into a new space can improve performance. One of the major problems with ST, is that the algorithm is O(n2m4), where n is the number of time series and m is the length of the series. As a problem increases in sizes, or additional dimensions are added, the algorithm quickly becomes computationally infeasible. The research question addressed is whether the shapelet transform be improved in terms of accuracy and speed. Making algorithmic improvements to shapelets will enable the development of multivariate shapelet algorithms that can attempt to solve much larger problems in realistic time frames. In support of this thesis a new distance early abandon method is proposed. A class balancing algorithm is implemented, which uses a one vs. all multi class information gain that enables heuristics which were developed for two class problems. To support these improvements a large scale analysis of the best shapelet algorithms is conducted as part of a larger experimental evaluation. ST is proven to be one of the most accurate algorithms in TSC on the UCR-UEA datasets. Contract classification is proposed for shapelets, where a fixed run time is set, and the number of shapelets is bounded. Four search algorithms are evaluated with fixed run times of one hour and one day, three of which are not significantly worse than a full enumeration. Finally, three multivariate shapelet algorithms are developed and compared to benchmark results and multivariate dynamic time warping.
18
Lintonen, T. (Timo). "Optimization in semi-supervised classification of multivariate time series". Master's thesis, University of Oulu, 2019. http://jultika.oulu.fi/Record/nbnfioulu-201902121196.
Texto completo da fonteResumo:
Abstract. In this thesis, I study methods that classify time series in a semi-supervised manner. I compare the performance of models that assume independent and identically distributed observations against models that assume nearby observations to be dependent of each other. These models are evaluated on three real world time series data sets. In addition, I carefully go through the theory of mathematical optimization behind two successful algorithms used in this thesis: Support Vector Data Description and Dynamic Time Warping. For the algorithm Dynamic Time Warping, I provide a novel proof that is based on dynamic optimization.
The experiments in this thesis suggest that the assumption of observations in time series to be independent and identically distributed may deteriorate the results of semi-supervised classification. The novel self-training method presented in this thesis called Peak Evaluation using Perceptually Important Points shows great performance on multivariate time series compared to the methods currently existing in literature. The feature subset selection of multivariate time series may improve classification performance, but finding a reliable unsupervised feature subset selection method remains an open question.
19
Булах, В. А., Л. О. Кіріченко e Т. А. Радівілова. "Classification of Multifractal Time Series by Decision Tree Methods". Thesis, КНУ, 2018. http://openarchive.nure.ua/handle/document/5840.
Texto completo da fonteResumo:
The article considers classification task of model fractal time series
by the methods of machine learning. To classify the series, it is proposed to use
the meta algorithms based on decision trees. To modeling the fractal time series, binomial stochastic cascade processes are used. Classification of time series by the ensembles of decision trees models is carried out. The analysis indicates that the best results are obtained by the methods of bagging and random
forest which use regression trees.
20
Lopez, Farias Rodrigo. "Time series forecasting based on classification of dynamic patterns". Thesis, IMT Alti Studi Lucca, 2015. http://e-theses.imtlucca.it/187/1/Farias_phdthesis.pdf.
Texto completo da fonteResumo:
This thesis addresses the problem of designing short-term forecasting
models for water demand time series presenting
nonlinear behaviour difficult to be fitted with single linear
models. These behaviours can be identified and classified
to build specialised models for performing local predictions
given an estimated operational regime. Each behavior class is
seen as a forecasting operation mode that activates a forecasting
model. For this purpose we developed a general modular
framework with three different implementations: An implementation
of a Multi-Model predictor that works with Machine
Learning regressors, clustering algorithms, classification,
and function approximations with the objective of producing
accurate forecasts for short horizons. The second and
third implementations are hybrid algorithms that use qualitative
and quantitative information from time series. The
quantitative component contains the aggregated magnitude
of each period of time and the qualitative component contains
the patterns associated with modes. For the qualitative
component we used a low order Seasonal ARIMA model and
for the qualitative component a k-Nearest Neighbours that
predicts the next pattern used to distribute the aggregated
magnitude given by the Seasonal ARIMA. The third implementation
is based on the same architecture, assuming the
existence of an accurate activity calendar with a sequence of
working and rest days, related to the forecast patterns. This
scheme is extended with a nonlinear filter module for the prediction
of pattern mismatches.
21
CABRI, ALBERTO. "Quantum inspired approach for early classification of time series". Doctoral thesis, Università degli studi di Genova, 2020. http://hdl.handle.net/11567/991085.
Texto completo da fonteResumo:
Is it possible to apply some fundamental principles of quantum-computing to time series classification algorithms? This is the initial spark that became the research question I decided to chase at the very beginning of my PhD studies. The idea came accidentally after reading a note on the ability of entanglement to express the correlation between two particles, even far away from each other. The test problem was also at hand because I was investigating on possible algorithms for real time bot detection, a challenging problem at present day, by means of statistical approaches for sequential classification. The quantum inspired algorithm presented in this thesis stemmed as an evolution of the statistical method mentioned above: it is a novel approach to address binary and multinomial classification of an incoming data stream, inspired by the principles of Quantum Computing, in order to ensure the shortest decision time with high accuracy. The proposed approach exploits the analogy between the intrinsic correlation of two or more particles and the dependence of each item in a data stream with the preceding ones. Starting from the a-posteriori probability of each item to belong to a particular class, we can assign a Qubit state representing a combination of the aforesaid probabilities for all available observations of the time series. By leveraging superposition and entanglement on subsequences of growing length, it is possible to devise a measure of membership to each class, thus enabling the system to take a reliable decision when a sufficient level of confidence is met. In order to provide an extensive and thorough analysis of the problem, a well-fitting approach for bot detection was replicated on our dataset and later compared with the statistical algorithm to determine the best option. The winner was subsequently examined against the new quantum-inspired proposal, showing the superior capability of the latter in both binary and multinomial classification of data streams. The validation of quantum-inspired approach in a synthetically generated use case, completes the research framework and opens new perspectives in on-the-fly time series classification, that we have just started to explore. Just to name a few ones, the algorithm is currently being tested with encouraging results in predictive maintenance and prognostics for automotive, in collaboration with University of Bradford (UK), and in action recognition from video streams.
22
Renard, Xavier. "Time series representation for classification : a motif-based approach". Thesis, Paris 6, 2017. http://www.theses.fr/2017PA066593/document.
Texto completo da fonteResumo:
Nos travaux décrits dans cette thèse portent sur l’apprentissage d’une représentation pour la classification automatique basée sur la découverte de motifs à partir de séries temporelles. L’information pertinente contenue dans une série temporelle peut être encodée temporellement sous forme de tendances, de formes ou de sous-séquences contenant habituellement des distorsions. Des approches ont été développées pour résoudre ces problèmes souvent au prix d’une importante complexité calculatoire. Parmi ces techniques nous pouvons citer les mesures de distance et les représentations de l’information contenue dans les séries temporelles. Nous nous concentrons sur la représentation de l’information contenue dans les séries temporelles. Nous proposons un cadre (framework) pour générer une nouvelle représentation de séries temporelles basée sur la découverte automatique d’ensembles discriminants de sous-séquences. Cette représentation est adaptée à l’utilisation d’algorithmes de classification classiques basés sur des attributs. Le framework proposé transforme un ensemble de séries temporelles en un espace d’attributs (feature space) à partir de sous-séquences énumérées des séries temporelles, de mesures de distance et de fonctions d’agrégation. Un cas particulier de ce framework est la méthode notoire des « shapelets ». L’inconvénient potentiel d’une telle approache est le nombre très important de sous-séquences à énumérer en ce qu’il induit un très grand feature space, accompagné d’une très grande complexité calculatoire. Nous montrons que la plupart des sous-séquences présentes dans un jeu de données composé de séries temporelles sont redondantes. De ce fait, un sous-échantillonnage aléatoire peut être utilisé pour générer un petit sous-ensemble de sous-séquences parmi l’ensemble exhaustif, en préservant l’information nécessaire pour la classification et tout en produisant un feature space de taille compatible avec l’utilisation d’algorithmes d’apprentissage automatique de l’état de l’art avec des temps de calculs raisonnable. On démontre également que le nombre de sous-séquences à tirer n’est pas lié avec le nombre de séries temporelles présent dans l’ensemble d’apprentissage, ce qui garantit le passage à l’échelle de notre approche. La combinaison de cette découverte dans le contexte de notre framework nous permet de profiter de techniques avancées (telles que des méthodes de sélection d’attributs multivariées) pour découvrir une représentation de séries temporelles plus riche, en prenant par exemple en considération les relations entre sous-séquences. Ces résultats théoriques ont été largement testés expérimentalement sur une centaine de jeux de données classiques de la littérature, composés de séries temporelles univariées et multivariées. De plus, nos recherches s’inscrivant dans le cadre d’une convention de recherche industrielle (CIFRE) avec Arcelormittal, nos travaux ont été appliqués à la détection de produits d’acier défectueux à partir des mesures effectuées par les capteurs sur des lignes de productionOur research described in this thesis is about the learning of a motif-based representation from time series to perform automatic classification. Meaningful information in time series can be encoded across time through trends, shapes or subsequences usually with distortions. Approaches have been developed to overcome these issues often paying the price of high computational complexity. Among these techniques, it is worth pointing out distance measures and time series representations. We focus on the representation of the information contained in the time series. We propose a framework to generate a new time series representation to perform classical feature-based classification based on the discovery of discriminant sets of time series subsequences (motifs). This framework proposes to transform a set of time series into a feature space, using subsequences enumerated from the time series, distance measures and aggregation functions. One particular instance of this framework is the well-known shapelet approach. The potential drawback of such an approach is the large number of subsequences to enumerate, inducing a very large feature space and a very high computational complexity. We show that most subsequences in a time series dataset are redundant. Therefore, a random sampling can be used to generate a very small fraction of the exhaustive set of subsequences, preserving the necessary information for classification and thus generating a much smaller feature space compatible with common machine learning algorithms with tractable computations. We also demonstrate that the number of subsequences to draw is not linked to the number of instances in the training set, which guarantees the scalability of the approach. The combination of the latter in the context of our framework enables us to take advantage of advanced techniques (such as multivariate feature selection techniques) to discover richer motif-based time series representations for classification, for example by taking into account the relationships between the subsequences. These theoretical results have been extensively tested on more than one hundred classical benchmarks of the literature with univariate and multivariate time series. Moreover, since this research has been conducted in the context of an industrial research agreement (CIFRE) with Arcelormittal, our work has been applied to the detection of defective steel products based on production line's sensor measurements
23
Brooks, Daniel. "Deep Learning and Information Geometry for Time-Series Classification". Electronic Thesis or Diss., Sorbonne université, 2020. http://www.theses.fr/2020SORUS276.
Texto completo da fonteResumo:
L’apprentissage automatique, et en particulier l’apprentissage profond, unit un arsenal d’outillages puissants pour modeler et étudier les distributions statistiques sous-jacentes aux données, permettant ainsi l’extraction d’informations sémantiquement valides et interprétables depuis des séquences tabulaires de nombres par ailleurs indigestes à l’œil humain. Bien que l’apprentissage fournisse une solution générique à la plupart des problèmes, certains types de données présentent une riche structure issue de phénomènes physiques: les images ont la localité spatiale, les sons la séquentialité temporelle, le radar la structure temps-fréquence. Il est à la fois intuitif et démontrable qu’il serait bénéfique d’exploiter avec astucieuse ces formations fondatrices au sein même des modèles d’apprentissage. A l’instar des architectures convolutives pour les images, les propriétés du signal peuvent être encodées et utilisées dans un réseau de neurones adapté, avec pour but l’apprentissage de modèles plus efficaces, plus performants. Spécifiquement, nous œuvrerons à intégrer dans la conception nos modèles profonds pour la classification de séries temporelles des sur leurs structures sous-jacentes, à savoir le temps, la fréquence, et leur nature proprement complexe. En allant plus loin dans une veine similaire, l’on peut s’atteler à la tâche d’étudier non pas le signal en tant que tel, mais bel et bien la distribution statistique dont il est issu. Dans ce scénario, les familles Gaussiennes constituent un candidat de choix. Formellement, la covariance des vecteurs de données caractérisent entièrement une telle distribution, pour peu qu’on la considère, à peu de frais, centrée; le développement d’algorithmes d’apprentissage, notamment profonds, sur des matrices de covariance, sera ainsi un thème central de cette thèse. L’espace des distributions diverge de manière fondamentale des espaces Euclidiens plats; il s’agit en fait de variétés Riemanniennes courbes, desquelles il conviendra de respecter la géométrie mathématique intrinsèque. Spécifiquement, nous contribuons à des architectures existantes par la création de nouvelles couches inspirées de la géométrie de l’information, notamment une couche de projection sensible aux données, et une couche inspirée de l’algorithme classique de la Batch Normalization. La validation empirique de nos nouveaux modèles se fera dans trois domaines différents: la reconnaissance d’émotions par vidéo, d’action par squelettes, avec une attention toute particulière à la classification de drones par signal radar micro-Doppler. Enfin, nous proposerons une librairie PyTorch aidant à la reproduction des résultats et la facilité de ré-implémentationdes algorithmes proposésMachine Learning, and in particular Deep Learning, is a powerful tool to model and study the intrinsic statistical foundations of data, allowing the extraction of meaningful, human-interpretable information from otherwise unpalatable arrays of floating points. While it provides a generic solution to many problems, some particular data types exhibit strong underlying physical structure: images have spatial locality, audio has temporal sequentiality, radar has time-frequency structure. Both intuitively and formally, there can be much to gain in leveraging this structure by adapting the subsequent learning models. As convolutional architectures for images, signal properties can be encoded and harnessed within the network. Conceptually, this would allow for a more intrinsic handling of the data, potentially leading to more efficient learning models. Thus, we will aim to use known structures in the signals as model priors. Specifically, we build dedicated deep temporal architectures for time series classification, and explore the use of complex values in neural networks to further refine the analysis of structured data. Going even further, one may wish to directly study the signal’s underlying statistical process. As such, Gaussian families constitute a popular candidate. Formally, the covariance of the data fully characterizes such a distribution; developing Machine Learning algorithms on covariance matrices will thus be a central theme throughout this thesis. Statistical distributions inherently diverge from the Euclidean framework; as such, it is necessary to study them on the appropriate, curved Riemannian manifold, as opposed to a flat, Euclidean space. Specifically, we contribute to existing deep architectures by adding normalizations in the form of data-aware mappings, and a Riemannian Batch Normalization algorithm. We showcase empirical validation through a variety of different tasks, including emotion and action recognition from video and Motion Capture data, with a sharpened focus on micro-Doppler radar data for Non-Cooperative Target Recognition drone recognition. Finally, we develop a library for the Deep Learning framework PyTorch, to spur reproducibility and ease of use
24
Renard, Xavier. "Time series representation for classification : a motif-based approach". Electronic Thesis or Diss., Paris 6, 2017. http://www.theses.fr/2017PA066593.
Texto completo da fonteResumo:
Nos travaux décrits dans cette thèse portent sur l’apprentissage d’une représentation pour la classification automatique basée sur la découverte de motifs à partir de séries temporelles. L’information pertinente contenue dans une série temporelle peut être encodée temporellement sous forme de tendances, de formes ou de sous-séquences contenant habituellement des distorsions. Des approches ont été développées pour résoudre ces problèmes souvent au prix d’une importante complexité calculatoire. Parmi ces techniques nous pouvons citer les mesures de distance et les représentations de l’information contenue dans les séries temporelles. Nous nous concentrons sur la représentation de l’information contenue dans les séries temporelles. Nous proposons un cadre (framework) pour générer une nouvelle représentation de séries temporelles basée sur la découverte automatique d’ensembles discriminants de sous-séquences. Cette représentation est adaptée à l’utilisation d’algorithmes de classification classiques basés sur des attributs. Le framework proposé transforme un ensemble de séries temporelles en un espace d’attributs (feature space) à partir de sous-séquences énumérées des séries temporelles, de mesures de distance et de fonctions d’agrégation. Un cas particulier de ce framework est la méthode notoire des « shapelets ». L’inconvénient potentiel d’une telle approache est le nombre très important de sous-séquences à énumérer en ce qu’il induit un très grand feature space, accompagné d’une très grande complexité calculatoire. Nous montrons que la plupart des sous-séquences présentes dans un jeu de données composé de séries temporelles sont redondantes. De ce fait, un sous-échantillonnage aléatoire peut être utilisé pour générer un petit sous-ensemble de sous-séquences parmi l’ensemble exhaustif, en préservant l’information nécessaire pour la classification et tout en produisant un feature space de taille compatible avec l’utilisation d’algorithmes d’apprentissage automatique de l’état de l’art avec des temps de calculs raisonnable. On démontre également que le nombre de sous-séquences à tirer n’est pas lié avec le nombre de séries temporelles présent dans l’ensemble d’apprentissage, ce qui garantit le passage à l’échelle de notre approche. La combinaison de cette découverte dans le contexte de notre framework nous permet de profiter de techniques avancées (telles que des méthodes de sélection d’attributs multivariées) pour découvrir une représentation de séries temporelles plus riche, en prenant par exemple en considération les relations entre sous-séquences. Ces résultats théoriques ont été largement testés expérimentalement sur une centaine de jeux de données classiques de la littérature, composés de séries temporelles univariées et multivariées. De plus, nos recherches s’inscrivant dans le cadre d’une convention de recherche industrielle (CIFRE) avec Arcelormittal, nos travaux ont été appliqués à la détection de produits d’acier défectueux à partir des mesures effectuées par les capteurs sur des lignes de productionOur research described in this thesis is about the learning of a motif-based representation from time series to perform automatic classification. Meaningful information in time series can be encoded across time through trends, shapes or subsequences usually with distortions. Approaches have been developed to overcome these issues often paying the price of high computational complexity. Among these techniques, it is worth pointing out distance measures and time series representations. We focus on the representation of the information contained in the time series. We propose a framework to generate a new time series representation to perform classical feature-based classification based on the discovery of discriminant sets of time series subsequences (motifs). This framework proposes to transform a set of time series into a feature space, using subsequences enumerated from the time series, distance measures and aggregation functions. One particular instance of this framework is the well-known shapelet approach. The potential drawback of such an approach is the large number of subsequences to enumerate, inducing a very large feature space and a very high computational complexity. We show that most subsequences in a time series dataset are redundant. Therefore, a random sampling can be used to generate a very small fraction of the exhaustive set of subsequences, preserving the necessary information for classification and thus generating a much smaller feature space compatible with common machine learning algorithms with tractable computations. We also demonstrate that the number of subsequences to draw is not linked to the number of instances in the training set, which guarantees the scalability of the approach. The combination of the latter in the context of our framework enables us to take advantage of advanced techniques (such as multivariate feature selection techniques) to discover richer motif-based time series representations for classification, for example by taking into account the relationships between the subsequences. These theoretical results have been extensively tested on more than one hundred classical benchmarks of the literature with univariate and multivariate time series. Moreover, since this research has been conducted in the context of an industrial research agreement (CIFRE) with Arcelormittal, our work has been applied to the detection of defective steel products based on production line's sensor measurements
25
Fulcher, Benjamin D. "Highly comparative time-series analysis". Thesis, University of Oxford, 2012. http://ora.ox.ac.uk/objects/uuid:642b65cf-4686-4709-9f9d-135e73cfe12e.
Texto completo da fonteResumo:
In this thesis, a highly comparative framework for time-series analysis is developed. The approach draws on large, interdisciplinary collections of over 9000 time-series analysis methods, or operations, and over 30 000 time series, which we have assembled. Statistical learning methods were used to analyze structure in the set of operations applied to the time series, allowing us to relate different types of scientific methods to one another, and to investigate redundancy across them. An analogous process applied to the data allowed different types of time series to be linked based on their properties, and in particular to connect time series generated by theoretical models with those measured from relevant real-world systems. In the remainder of the thesis, methods for addressing specific problems in time-series analysis are presented that use our diverse collection of operations to represent time series in terms of their measured properties. The broad utility of this highly comparative approach is demonstrated using various case studies, including the discrimination of pathological heart beat series, classification of Parkinsonian phonemes, estimation of the scaling exponent of self-affine time series, prediction of cord pH from fetal heart rates recorded during labor, and the assignment of emotional content to speech recordings. Our methods are also applied to labeled datasets of short time-series patterns studied in temporal data mining, where our feature-based approach exhibits benefits over conventional time-domain classifiers. Lastly, a feature-based dimensionality reduction framework is developed that links dependencies measured between operations to the number of free parameters in a time-series model that could be used to generate a time-series dataset.
26
Nilsson, Markus. "A case-based approach for classification of physiological time-series /". Västerås : Mälardalen University, 2004. http://www.mrtc.mdh.se/publications/0718.pdf.
Texto completo da fonte
27
Lundkvist, Emil. "Decision Tree Classification and Forecasting of Pricing Time Series Data". Thesis, KTH, Reglerteknik, 2014. http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-151017.
Texto completo da fonteResumo:
Many companies today, in different fields of operations and sizes, have access to a vast amount of data which was not available only a couple of years ago. This situation gives rise to questions regarding how to organize and use the data in the best way possible. In this thesis a large database of pricing data for products within various market segments is analysed. The pricing data is from both external and internal sources and is therefore confidential. Because of the confidentiality, the labels from the database are in this thesis substituted with generic ones and the company is not referred to by name, but the analysis is carried out on the real data set. The data is from the beginning unstructured and difficult to overlook. Therefore, it is first classified. This is performed by feeding some manual training data into an algorithm which builds a decision tree. The decision tree is used to divide the rest of the products in the database into classes. Then, for each class, a multivariate time series model is built and each product’s future price within the class can be predicted. In order to interact with the classification and price prediction, a front end is also developed. The results show that the classification algorithm both is fast enough to operate in real time and performs well. The time series analysis shows that it is possible to use the information within each class to do predictions, and a simple vector autoregressive model used to perform it shows good predictive results.
28
Phan, Thi-Thu-Hong. "Elastic matching for classification and modelisation of incomplete time series". Thesis, Littoral, 2018. http://www.theses.fr/2018DUNK0483/document.
Texto completo da fonteResumo:
Les données manquantes constituent un challenge commun en reconnaissance de forme et traitement de signal. Une grande partie des techniques actuelles de ces domaines ne gère pas l'absence de données et devient inutilisable face à des jeux incomplets. L'absence de données conduit aussi à une perte d'information, des difficultés à interpréter correctement le reste des données présentes et des résultats biaisés notamment avec de larges sous-séquences absentes. Ainsi, ce travail de thèse se focalise sur la complétion de larges séquences manquantes dans les séries monovariées puis multivariées peu ou faiblement corrélées. Un premier axe de travail a été une recherche d'une requête similaire à la fenêtre englobant (avant/après) le trou. Cette approche est basée sur une comparaison de signaux à partir d'un algorithme d'extraction de caractéristiques géométriques (formes) et d'une mesure d'appariement élastique (DTW - Dynamic Time Warping). Un package R CRAN a été développé, DTWBI pour la complétion de série monovariée et DTWUMI pour des séries multidimensionnelles dont les signaux sont non ou faiblement corrélés. Ces deux approches ont été comparées aux approches classiques et récentes de la littérature et ont montré leur faculté de respecter la forme et la dynamique du signal. Concernant les signaux peu ou pas corrélés, un package DTWUMI a aussi été développé. Le second axe a été de construire une similarité floue capable de prender en compte les incertitudes de formes et d'amplitude du signal. Le système FSMUMI proposé est basé sur une combinaison floue de similarités classiques et un ensemble de règles floues. Ces approches ont été appliquées à des données marines et météorologiques dans plusieurs contextes : classification supervisée de cytogrammes phytoplanctoniques, segmentation non supervisée en états environnementaux d'un jeu de 19 capteurs issus d'une station marine MAREL CARNOT en France et la prédiction météorologique de données collectées au VietnamMissing data are a prevalent problem in many domains of pattern recognition and signal processing. Most of the existing techniques in the literature suffer from one major drawback, which is their inability to process incomplete datasets. Missing data produce a loss of information and thus yield inaccurate data interpretation, biased results or unreliable analysis, especially for large missing sub-sequence(s). So, this thesis focuses on dealing with large consecutive missing values in univariate and low/un-correlated multivariate time series. We begin by investigating an imputation method to overcome these issues in univariate time series. This approach is based on the combination of shape-feature extraction algorithm and Dynamic Time Warping method. A new R-package, namely DTWBI, is then developed. In the following work, the DTWBI approach is extended to complete large successive missing data in low/un-correlated multivariate time series (called DTWUMI) and a DTWUMI R-package is also established. The key of these two proposed methods is that using the elastic matching to retrieving similar values in the series before and/or after the missing values. This optimizes as much as possible the dynamics and shape of knowledge data, and while applying the shape-feature extraction algorithm allows to reduce the computing time. Successively, we introduce a new method for filling large successive missing values in low/un-correlated multivariate time series, namely FSMUMI, which enables to manage a high level of uncertainty. In this way, we propose to use a novel fuzzy grades of basic similarity measures and fuzzy logic rules. Finally, we employ the DTWBI to (i) complete the MAREL Carnot dataset and then we perform a detection of rare/extreme events in this database (ii) forecast various meteorological univariate time series collected in Vietnam
29
LI, YUANXUN. "SVM Object Based Classification Using Dense Satellite Imagery Time Series". Thesis, KTH, Geoinformatik, 2018. http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-233340.
Texto completo da fonte
30
Hartvigsen, Thomas. "Adaptively-Halting RNN for Tunable Early Classification of Time Series". Digital WPI, 2018. https://digitalcommons.wpi.edu/etd-theses/1257.
Texto completo da fonteResumo:
Early time series classification is the task of predicting the class label of a time series before it is observed in its entirety. In time-sensitive domains where information is collected over time it is worth sacrificing some classification accuracy in favor of earlier predictions, ideally early enough for actions to be taken. However, since accuracy and earliness are contradictory objectives, a solution to this problem must find a task-dependent trade-off. There are two common state-of-the-art methods. The first involves an analyst selecting a timestep at which all predictions must be made. This does not capture earliness on a case-by-case basis, so if the selecting timestep is too early, all later signals are missed, and if a signal happens early, the classifier still waits to generate a prediction. The second method is the exhaustive search for signals, which encodes no timing information and is not scalable to high dimensions or long time series. We design the first early classification model called EARLIEST to tackle this multi-objective optimization problem, jointly learning (1) to decide at which time step to halt and generate predictions and (2) how to classify the time series. Each of these is learned based on the task and data features. We achieve an analyst-controlled balance between the goals of earliness and accuracy by pairing a recurrent neural network that learns to classify time series as a supervised learning task with a stochastic controller network that learns a halting-policy as a reinforcement learning task. The halting-policy dictates sequential decisions, one per timestep, of whether or not to halt the recurrent neural network and classify the time series early. This pairing of networks optimizes a global objective function that incorporates both earliness and accuracy. We validate our method via critical clinical prediction tasks in the MIMIC III database from the Beth Israel Deaconess Medical Center along with another publicly available time series classification dataset. We show that EARLIEST out-performs two state-of-the-art LSTM-based early classification methods. Additionally, we dig deeper into our model's performance using a synthetic dataset which shows that EARLIEST learns to halt when it observes signals without having explicit access to signal locations. The contributions of this work are three-fold. First, our method is the first neural network-based solution to early classification of time series, bringing the recent successes of deep learning to this problem. Second, we present the first reinforcement-learning based solution to the unsupervised nature of early classification, learning the underlying distributions of signals without access to this information through trial and error. Third, we propose the first joint-optimization of earliness and accuracy, allowing learning of complex relationships between these contradictory goals.
31
Caiado, Aníbal Jorge da Costa Cristóvão. "Distance-based methods for classification and clustering of time series". Doctoral thesis, Instituto Superior de Economia e Gestão, 2006. http://hdl.handle.net/10400.5/3531.
Texto completo da fonte
32
Mao, Dong. "Biological time series classification via reproducing kernels and sample entropy". Related electronic resource: Current Research at SU : database of SU dissertations, recent titles available full text, 2008. http://wwwlib.umi.com/cr/syr/main.
Texto completo da fonte
33
Zheng, Yue Chu. "Feature extraction for chart pattern classification in financial time series". Thesis, University of Macau, 2018. http://umaclib3.umac.mo/record=b3950623.
Texto completo da fonte
34
Granberg, Patrick. "Churn prediction using time series data". Thesis, KTH, Skolan för elektroteknik och datavetenskap (EECS), 2020. http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-294206.
Texto completo da fonteResumo:
Customer churn is problematic for any business trying to expand their customer base. The acquisition of new customers to replace churned ones are associated with additional costs, whereas taking measures to retain existing customers may prove more cost efficient. As such, it is of interest to estimate the time until the occurrence of a potential churn for every customer in order to take preventive measures. The application of deep learning and machine learning to this type of problem using time series data is relatively new and there is a lot of recent research on this topic. This thesis is based on the assumption that early signs of churn can be detected by the temporal changes in customer behavior. Recurrent neural networks and more specifically long short-term memory (LSTM) and gated recurrent unit (GRU) are suitable contenders since they are designed to take the sequential time aspect of the data into account. Random forest (RF) and stochastic vector machine (SVM) are machine learning models that are frequently used in related research. The problem is solved through a classification approach, and a comparison is done with implementations using LSTM, GRU, RF, and SVM. According to the results, LSTM and GRU perform similarly while being slightly better than RF and SVM in the task of predicting customers that will churn in the coming six months, and that all models could potentially lead to cost savings according to simulations (using non-official but reasonable costs assigned to each prediction outcome). Predicting the time until churn is a more difficult problem and none of the models can give reliable estimates, but all models are significantly better than random predictions.Kundbortfall är problematiskt för företag som försöker expandera sin kundbas. Förvärvandet av nya kunder för att ersätta förlorade kunder är associerat med extra kostnader, medan vidtagandet av åtgärder för att behålla kunder kan visa sig mer lönsamt. Som så är det av intresse att för varje kund ha pålitliga tidsestimat till en potentiell uppsägning kan tänkas inträffa så att förebyggande åtgärder kan vidtas. Applicerandet av djupinlärning och maskininlärning på denna typ av problem som involverar tidsseriedata är relativt nytt och det finns mycket ny forskning kring ämnet. Denna uppsats är baserad på antagandet att tidiga tecken på kundbortfall kan upptäckas genom kunders användarmönster över tid. Reccurent neural networks och mer specifikt long short-term memory (LSTM) och gated recurrent unit (GRU) är lämpliga modellval eftersom de är designade att ta hänsyn till den sekventiella tidsaspekten i tidsseriedata. Random forest (RF) och stochastic vector machine (SVM) är maskininlärningsmodeller som ofta används i relaterad forskning. Problemet löses genom en klassificeringsapproach, och en jämförelse utförs med implementationer av LSTM, GRU, RF och SVM. Resultaten visar att LSTM och GRU presterar likvärdigt samtidigt som de presterar bättre än RF och SVM på problemet om att förutspå kunder som kommer att säga upp sig inom det kommande halvåret, och att samtliga modeller potentiellt kan leda till kostnadsbesparingar enligt simuleringar (som använder icke-officiella men rimliga kostnader associerat till varje utfall). Att förutspå tid till en kunduppsägning är ett svårare problem och ingen av de framtagna modellerna kan ge pålitliga tidsestimat, men alla är signifikant bättre än slumpvisa gissningar.
35
Ribeiro, Joana Patrícia Bordonhos. "Outlier identification in multivariate time series". Master's thesis, Universidade de Aveiro, 2017. http://hdl.handle.net/10773/22200.
Texto completo da fonteResumo:
Mestrado em Matemática e AplicaçõesCom o desenvolvimento tecnológico, existe uma cada vez maior disponibilidade de dados. Geralmente representativos de situações do dia-a-dia, a existência de grandes quantidades de informação tem o seu interesse quando permite a extração de valor para o mercado. Além disso, surge importância em analisar não só os valores disponíveis mas também a sua associação com o tempo. A existência de valores anormais é inevitável. Geralmente denotados como outliers, a procura por estes valores é realizada comummente com o intuito de fazer a sua exclusão do estudo. No entanto, os outliers representam muitas vezes um objetivo de estudo. Por exemplo, no caso de deteção de fraudes bancárias ou no diagnóstico de doenças, o objetivo central é identificar situações anormais. Ao longo desta dissertação é apresentada uma metodologia que permite detetar outliers em séries temporais multivariadas, após aplicação de métodos de classificação. A abordagem escolhida é depois aplicada a um conjunto de dados real, representativo do funcionamento de caldeiras. O principal objetivo é identificar as suas falhas. Consequentemente, pretende-se melhorar os componentes do equipamento e portanto diminuir as suas falhas. Os algoritmos implementados permitem identificar não só as falhas do aparelho mas também o seu funcionamento normal. Pretende-se que as metodologias escolhidas sejam também aplicadas nos aparelhos futuros, permitindo melhorar a identificação em tempo real das falhas.
With the technological development, there is an increasing availability of data. Usually representative of day-to-day actions, the existence of large amounts of information has its own interest when it allows to extract value to the market. In addition, it is important to analyze not only the available values but also their association with time. The existence of abnormal values is inevitable. Usually denoted as outliers, the search for these values is commonly made in order to exclude them from the study. However, outliers often represent a goal of study. For example, in the case of bank fraud detection or disease diagnosis, the central objective is to identify the abnormal situations. Throughout this dissertation we present a methodology that allows the detection of outliers in multivariate time series, after application of classification methods. The chosen approach is then applied to a real data set, representative of boiler operation. The main goal is to identify faults. It is intended to improve boiler components and, hence, reduce the faults. The implemented algorithms allow to identify not only the boiler faults but also their normal operation cycles. We aim that the chosen methodologies will also be applied in future devices, allowing to improve real-time fault identification.
36
Leverger, Colin. "Investigation of a framework for seasonal time series forecasting". Thesis, Rennes 1, 2020. http://www.theses.fr/2020REN1S033.
Texto completo da fonteResumo:
Pour déployer des applications web, l'utilisation de serveurs informatique est primordiale. S'ils sont peu nombreux, les performances des applications peuvent se détériorer. En revanche, s'ils sont trop nombreux, les ressources sont gaspillées et les coûts argumentés. Dans ce contexte, les ingénieurs utilisent des outils de planning capacitaire qui leur permettent de suivre les performances des serveurs, de collecter les données temporelles générées par les infrastructures et d’anticiper les futurs besoins. La nécessité de créer des prévisions fiables apparaît évidente. Les données des infrastructures présentent souvent une saisonnalité évidente. Le cycle d’activité suivi par l’infrastructure est déterminé par certains cycles saisonniers (par exemple, le rythme quotidien de l’activité des utilisateurs). Cette thèse présente un framework pour la prévision de séries temporelles saisonnières. Ce framework est composé de deux modèles d’apprentissage automatique (e.g. clustering et classification) et vise à fournir des prévisions fiables à moyen terme avec un nombre limité de paramètres. Trois implémentations du framework sont présentées : une baseline, une déterministe et une probabiliste. La baseline est constituée d'un algorithme de clustering K-means et de modèles de Markov. La version déterministe est constituée de plusieurs algorithmes de clustering (K-means, K-shape, GAK et MODL) et de plusieurs classifieurs (classifieurs bayésiens, arbres de décisions, forêt aléatoire et régression logistique). La version probabiliste repose sur du coclustering pour créer des grilles probabilistes de séries temporelles, afin de décrire les données de manière non supervisée. Les performances des différentes implémentations du framework sont comparées avec différents modèles de l’état de l’art, incluant les modèles autorégressifs, les modèles ARIMA et SARIMA, les modèles Holts Winters, ou encore Prophet pour la partie probabiliste. Les résultats de la baseline sont encourageants, et confirment l'intérêt pour le framework proposé. De bons résultats sont constatés pour la version déterministe du framework, et des résultats corrects pour la version probabiliste. Un cas d’utilisation d’Orange est étudié, et l’intérêt et les limites de la méthodologie sont montrésTo deploy web applications, using web servers is paramount. If there is too few of them, applications performances can quickly deteriorate. However, if they are too numerous, the resources are wasted and the cost increased. In this context, engineers use capacity planning tools to follow the performances of the servers, to collect time series data and to anticipate future needs. The necessity to create reliable forecasts seems clear. Data generated by the infrastructure often exhibit seasonality. The activity cycle followed by the infrastructure is determined by some seasonal cycles (for example, the user’s daily rhythms). This thesis introduces a framework for seasonal time series forecasting. This framework is composed of two machine learning models (e.g. clustering and classification) and aims at producing reliable midterm forecasts with a limited number of parameters. Three instantiations of the framework are presented: one baseline, one deterministic and one probabilistic. The baseline is composed of K-means clustering algorithms and Markov Models. The deterministic version is composed of several clustering algorithms (K-means, K-shape, GAK and MODL) and of several classifiers (naive-bayes, decision trees, random forests and logistic regression). The probabilistic version relies on coclustering to create time series probabilistic grids, that are used to describe the data in an unsupervised way. The performances of the various implementations are compared with several state-of-the-art models, including the autoregressive models, ARIMA and SARIMA, Holt Winters, or even Prophet for the probabilistic paradigm. The results of the baseline are encouraging and confirm the interest for the framework proposed. Good results are observed for the deterministic implementation, and correct results for the probabilistic version. One Orange use case is studied, and the interest and limits of the methodology are discussed
37
Sengupta, Shreeya. "Multidimensional time series classification and its application to video activity recognition". Thesis, Ulster University, 2015. http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.673808.
Texto completo da fonteResumo:
A collection of observations made sequentially through time is known as a time series. The order in which the observations in a time series are recorded is important and is a distinct characteristic of time series. A time series may have only one dimension (unidimensional) or may have multiple dimensions (multidimensional). The research presented in this thesis considers multidimensional time series. Financial stock market, videos, medical (EEG and ECG) and speech data are all examples of multidimensional time series data. Analysis of multidimensional time series data can reveal underlying patterns, the knowledge of which can benefit several time series applications. For example, rules derived by analysing the stock data can be helpful in predicting the behaviour of the stock market and identifying the pattern of the strokes in signatures can aid signature verification. However, time series analysis is often hindered by the presence of variability in the series. Variability refers to the difference in the time series data generated at different points of time. It arises because of the stochasticity of the process generating the time series, non-stationarity of time series, presence of noise in time series and the variable sampling rate with which a time series is sampled. This research studies the effect of non-stationarity and variability on multidimensional time series analysis, with a pat1icular focus on video activity recognition. The research, firstly, studies the effect of non-stationarity, one of the causes of variability, and variability on time series analysis in general. The efficacy of several analysis models was evaluated on various time series problems. Results show that both non-stationarity and variability degrades the performance of the models consequently affecting time series analysis. Then, the research concentrates on video data analysis where space and time variabilities abound. The variability of the video content along the space and time dimension is known as spatial and temporal variability respectively. New methods to handle / minimise the effect of spatial and temporal variabilities are proposed. The proposed methods are then assessed against existing methods which do not handle the spatial and temporal variabilities explicitly. The proposed methods perform better than the existing methods, indicating that better video activity recognition can be achieved by addressing the spatial and temporal variabilities
38
Wheeler, Brandon Myles. "Evaluating time-series smoothing algorithms for multi-temporal land cover classification". Thesis, Virginia Tech, 2015. http://hdl.handle.net/10919/74313.
Texto completo da fonteResumo:
In this study we applied the asymmetric Gaussian, double-logistic, and Savitzky-Golay filters to
MODIS time-series NDVI data to compare the capability of smoothing algorithms in noise
reduction for improving land cover classification in the Great Lakes Basin, and providing
groundwork to support cyanobacteria and cyanotoxin monitoring efforts. We used inter-class
separability and intra-class variability, at varying levels of pixel homogeneity, to evaluate the
effectiveness of three smoothing algorithms. Based on these initial tests, the algorithm which
returned the best results was used to analyze how image stratification by ecoregion can affect
filter performance.
MODIS 16-day 250m NDVI imagery of the Great Lakes Basin from 2001-2013 were used in
conjunction with National Land Cover Database (NLCD) 2006 and 2011 data, and Cropland Data
Layers (CDL) from 2008 to 2013 to conduct these evaluations. Inter-class separability was
measured by Jeffries-Matusita (JM) distances between selected land cover classes (both general classes and specific crops), and intra-class variability was measured by calculating simple
Euclidean distance for samples within a land cover class. Within the study area, it was found that the application of a smoothing algorithm can significantly reduce image noise, improving both inter-class separability and intra-class variability when compared to the raw data. Of the three
filters examined, the asymmetric Gaussian filter consistently returned the highest values of interclass separability, while all three filters performed very similarly for within-class variability. The ecoregion analysis based on the asymmetric Gaussian dataset indicated that the scale of study area can heavily impact within-class separability. The criteria we established have potential for furthering our understanding of the strengths and weaknesses of different smoothing algorithms, thereby improving pre-processing decisions for land cover classification using time-series data.Master of Science
39
Rezvanizaniani, Seyed Mohammad. "Probabilistic Based Classification Techniques for Improved Prognostics Using Time Series Data". University of Cincinnati / OhioLINK, 2015. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1428048932.
Texto completo da fonte
40
Pienaar, Marc. "On the classification of time series and cross wavelet phase variance". Doctoral thesis, University of Cape Town, 2016. http://hdl.handle.net/11427/22869.
Texto completo da fonteResumo:
The continuous wavelet transform (CWT) is arguably one of the best tools to explore underlying characteristic features of time series data. Its application in large time series classification experiments, however, has been severely limited due to the large amount of redundant associated information. By extending the capabilities of the CWT to perform cross wavelet analysis (CWA), common frequency behaviour between two time series is highlighted, and the potential to extract amplitude modulated (AM) and frequency modulation (FM) characteristics in an automated way is possible. Characterisation of AM is relatively straightforward and can be resolved by any number of Euclidean based techniques in both the time and frequency domains. FM on the other hand, is somewhat more difficult as it transcends multiple wavelet scales. In this study, linear combinations of scales are used to extract both AM similarity (derived from global wavelet power spectra) and FM coherency, using a new method developed called cross wavelet phase variance (CWPV). The methodology is applied to large scale classification problems (using benchmark time series), in which the method clearly outperforms other common distance-based measures. Lastly, the approach provides a powerful framework in which AM and FM characteristics common between time series can be explicitly mapped to their corresponding scales, and with some initial optimisation, can be applied to any classification problem.
41
Kaffashi, Farhad. "Variability analysis & its applications to physiological time series data". online version, 2007. http://rave.ohiolink.edu/etdc/view?acc%5Fnum=case1181072302.
Texto completo da fonte
42
Dietrich, Christian R. [Verfasser]. "Temporal sensorfusion for the classification of bioacoustic time series / Christian R. Dietrich". Ulm : Universität Ulm. Fakultät für Informatik, 2004. http://d-nb.info/1015438903/34.
Texto completo da fonte
43
Ridnert, Carl. "Machine Learning for Sparse Time-Series Classification - An Application in Smart Metering". Thesis, KTH, Matematisk statistik, 2019. http://urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-243925.
Texto completo da fonteResumo:
Smart Meters are measuring devices collecting labeled time series data of utility consumptions from sub-meters and are capable of automatically transmit-ting this between the customer and utility companies together with other companies that offer services such as monitoring of consumption and cleaning of data. The smart meters are in some cases experiencing communication errors. One such error occurs when the information about what the utility sub-meters are measuring is lost. This information is important for when the producers of the utility are billing the customers for their usage. The information has had to be collected manually, something which is inefficient in terms of time and money. In this thesis a method for classifying the meters based on their raw time series data is investigated. The data used in the thesis comes from Metry AB and contains thousands of time series in five different classes. The task is complicated by the fact that the data has a high class imbalance, contains many missing values and that the time series vary substantially in length. The proposed method is based on partitioning the time series into slices of equal size and training a Deep Neural Network (DNN) together with a Bayesian Neural Network (BNN) to classify the slices. Prediction on new time series is performed by the prediction of individual slices for that time series followed by a voting procedure. The method is justified through a set of assumptions about the underlying stochastic process generating the time series coupled with an analysis based on the multinomial distribution. The results indicate that the models tend to perform worse on the samples coming from the classes ”water” and ”hot water” and that the worst performance is on the ”hot water”-class. On all the classes the models achieve accuracies of around 60%, by excluding the ”hot water” class it is possible to achieve accuracies of at least 70% on the data set. The models perform worse on time series that contain a few number of good quality slices, by considering only time series which has many good quality slices, accuracies of 70% are achieved for all classes and above 80% when excluding ”Hot Water”. It is concluded that in order to further improve the classification performance, more data is needed. Drawbacks with the method are the increased number of hyper-parameters involved in the extraction of slices. However, the voting method seems promising enough to investigate further on more highly sparse data sets.Smarta Mätare är maskiner kapabla att automatiskt sända data från sub-mätare mätandes förbrukningar av nyttigheter(utility) mellan kunden och företag som producerar nyttigheterna. Detta har inneburit att en marknad har öppnats upp för företag som tar förbrukningsdata och erbjuder tjänster så som appar där kunden kan se sin förbrukning samt rensning eller interpolering av data. Denna kommunikation har inneburit vissa problem, ett identifierat sådant är att det händer att information om vilken nyttighet som har mätts går förlorat. Denna information är viktig och har tidigare behövt hämtas manuellt på ett eller annat sätt, något som är ineffektivt. I detta examensarbete undersöks huruvida den informationen går att få tag på med enbart rådatan och klassificeringsalgoritmer. Datan kommer från Metry AB och innehåller tusentals tidsserier från fem olika klasser. Uppgiften försvåras av att datan uppvisar en stor obalans i klasserna, innehåller många saknade datapunkter och att tidsserierna varierar stort i längd. Metoden som föreslås baseras på en uppstyckning av tidsserierna i så kallade ”slices” av samma storlek och att träna Djupa Neurala Nätverk (DNN) och Bayesiska Neurala Nätverk (BNN) på dessa. Klassificering av nya tidsserier sker genom att låta modellerna rösta på slices från dem och välja den klass som får flest röster. Arbetet innehåller en teoretisk analys av röstningsprocessen baserat på en multinomial fördelning kombinerat med olika antaganden om processen som genererar dessa slices, denna syftar till att motivera valet av metod. Resultaten visar att modellerna kan tränas och korrekt klassificera mätarna till en viss grad samt att röstningsprocessen tenderar till att ge bättre resultat än att bara använda en slice per mätare. Det påvisas att prestandan är mycket sämre för en specifik klass, genom att exkludera den klassen så lyckas modellerna prestera slutgiltiga noggrannheter på mellan 70 − 80%. Det påvisas vissa skillnader mellan BNN modellen och DNN modellen i termer av noggrannhet, dock så är skillnaderna för små för att det ska gå att dra några generella slutsatser om vilken klassificeringsalgoritm som är bäst. Slutsatserna är att metoden verkar fungera rimligt väl på denna typ av data men att det behövs mer arbete för att förstå när den fungerar och hur man kan göra den bättre, detta är framtida arbete. Den största möjligheten till förbättring för just denna tillämpning identifieras vara att samla in mer data.
44
Wang, Zhihao. "Land Cover Classification on Satellite Image Time Series Using Deep Learning Models". The Ohio State University, 2020. http://rave.ohiolink.edu/etdc/view?acc_num=osu159559249009195.
Texto completo da fonte
45
Thungtong, Anurak. "Synchronization, Variability, and Nonlinearity Analysis: Applications to Physiological Time Series". Case Western Reserve University School of Graduate Studies / OhioLINK, 2013. http://rave.ohiolink.edu/etdc/view?acc_num=case1364316597.
Texto completo da fonte
46
Zoltan, Geler. "Role of Similarity Measures in Time Series Analysis". Phd thesis, Univerzitet u Novom Sadu, Prirodno-matematički fakultet u Novom Sadu, 2015. https://www.cris.uns.ac.rs/record.jsf?recordId=94848&source=NDLTD&language=en.
Texto completo da fonteResumo:
The subject of this dissertation encompasses a comprehensive overviewand analysis of the impact of Sakoe-Chiba global constraint on the mostcommonly used elastic similarity measures in the field of time-series datamining with a focus on classification accuracy. The choice of similaritymeasure is one of the most significant aspects of time-series analysis - itshould correctly reflect the resemblance between the data presented inthe form of time series. Similarity measures represent a criticalcomponent of many tasks of mining time series, including: classification,clustering, prediction, anomaly detection, and others.The research covered by this dissertation is oriented on several issues:1. review of the effects of global constraints on theperformance of computing similarity measures,2. a detailed analysis of the influence of constraining the elasticsimilarity measures on the accuracy of classical classificationtechniques,3. an extensive study of the impact of different weightingschemes on the classification of time series,4. development of an open source library that integrates themain techniques and methods required for analysis andmining time series, and which is used for the realization ofthese experimentsPredmet istraživanja ove disertacije obuhvata detaljan pregled i analizu uticaja Sakoe-Chiba globalnog ograničenja na najčešće korišćene elastične mere sličnosti u oblasti data mining-a vremenskih serija sa naglaskom na tačnost klasifikacije. Izbor mere sličnosti jedan je od najvažnijih aspekata analize vremenskih serija - ona treba verno reflektovati sličnost između podataka prikazanih u obliku vremenskih serija. Mera sličnosti predstavlјa kritičnu komponentu mnogih zadataka mining-a vremenskih serija, uklјučujući klasifikaciju, grupisanje (eng. clustering), predviđanje, otkrivanje anomalija i drugih.Istraživanje obuhvaćeno ovom disertacijom usmereno je na nekoliko pravaca:1. pregled efekata globalnih ograničenja na performanse računanja mera sličnosti,2. detalјna analiza posledice ograničenja elastičnih mera sličnosti na tačnost klasifikacije klasičnih tehnika klasifikacije,3. opsežna studija uticaj različitih načina računanja težina (eng. weighting scheme) na klasifikaciju vremenskih serija,4. razvoj biblioteke otvorenog koda (Framework for Analysis and Prediction - FAP) koja će integrisati glavne tehnike i metode potrebne za analizu i mining vremenskih serija i koja je korišćena za realizaciju ovih eksperimenata.
Predmet istraživanja ove disertacije obuhvata detaljan pregled i analizu uticaja Sakoe-Chiba globalnog ograničenja na najčešće korišćene elastične mere sličnosti u oblasti data mining-a vremenskih serija sa naglaskom na tačnost klasifikacije. Izbor mere sličnosti jedan je od najvažnijih aspekata analize vremenskih serija - ona treba verno reflektovati sličnost između podataka prikazanih u obliku vremenskih serija. Mera sličnosti predstavlja kritičnu komponentu mnogih zadataka mining-a vremenskih serija, uključujući klasifikaciju, grupisanje (eng. clustering), predviđanje, otkrivanje anomalija i drugih.Istraživanje obuhvaćeno ovom disertacijom usmereno je na nekoliko pravaca:1. pregled efekata globalnih ograničenja na performanse računanja mera sličnosti,2. detaljna analiza posledice ograničenja elastičnih mera sličnosti na tačnost klasifikacije klasičnih tehnika klasifikacije,3. opsežna studija uticaj različitih načina računanja težina (eng. weighting scheme) na klasifikaciju vremenskih serija,4. razvoj biblioteke otvorenog koda (Framework for Analysis and Prediction - FAP) koja će integrisati glavne tehnike i metode potrebne za analizu i mining vremenskih serija i koja je korišćena za realizaciju ovih eksperimenata.
47
Herbst, Gernot. "Online Recognition of Fuzzy Time Series Patterns". Universitätsbibliothek Chemnitz, 2009. http://nbn-resolving.de/urn:nbn:de:bsz:ch1-200901287.
Texto completo da fonteResumo:
This article deals with the recognition of recurring
multivariate time series patterns modelled sample-point-wise by parametric
fuzzy sets. An efficient classification-based approach for the
online recognition of incompleted developing patterns in streaming
time series is being presented. Furthermore, means are introduced
to enable users of the recognition system to restrict results to certain
stages of a pattern’s development, e. g. for forecasting purposes, all
in a consistently fuzzy manner.
48
Guo, Zhenyu 1963. "Time-frequency representation and pattern recognition of doppler blood flow signal for stenosis classification". Thesis, McGill University, 1993. http://digitool.Library.McGill.CA:80/R/?func=dbin-jump-full&object_id=41252.
Texto completo da fonteResumo:
The objective of this thesis is to characterize the time-frequency representation (TFR) of the Doppler blood flow signal in order to improve the detection and quantification of arterial stenoses. First of all, we studied the statistical behavior of the Doppler blood flow signal and found that the signal is Gaussian and inherently nonstationary. However, the usual assumption of stationarity during a short time interval ($ leq$10 ms) is acceptable. The signal was then modeled as a complex autoregressive (AR) process to provide directional information about the blood flow. It was shown that the TFR based on AR modeling is less sensitive to window length and sampling frequency than the spectrogram. In order to obtain a more precise Doppler TFR, a new distribution based on a Bessel kernel was proposed. This Bessel distribution can suppress the cross-terms effectively and has many desirable properties with high resolution in time and frequency. A numerical alias-free implementation of this distribution was also developed.A realistic model of the Doppler blood flow signal of the femoral artery was proposed and used to simulate signals for comparing five TFR techniques: the short-time Fourier transform, AR modeling, the Bessel distribution, the Choi-Williams distribution, and the Choi-Williams Reduced Interference Distribution. The results showed that the Bessel distribution is the best, but the Choi-Williams distribution and AR modeling are also good techniques to compute Doppler TFRs. The short-time Fourier transform, AR modeling, and the Bessel distribution were then applied to clinical data to derive diagnostic features for a pattern recognition system to assess lower limb arterial stenoses in 37 patients. A total of 379 arterial segments were classified into three stenotic classes. The results were in agreement with those based on computer-simulated signals, and confirmed that the Bessel distribution and AR modeling improve the Doppler spectral estimates and thus provide better classification of arterial stenosis.
49
Herbst, Gernot. "Short-Time Prediction Based on Recognition of Fuzzy Time Series Patterns". Universitätsbibliothek Chemnitz, 2010. http://nbn-resolving.de/urn:nbn:de:bsz:ch1-201001012.
Texto completo da fonteResumo:
This article proposes knowledge-based short-time prediction methods for multivariate streaming time series, relying on the early recognition of local patterns. A parametric, well-interpretable model for such patterns is presented, along with an online, classification-based recognition procedure. Subsequently, two options are discussed to predict time series employing the fuzzified pattern knowledge, accompanied by an example. Special emphasis is placed on comprehensible models and methods, as well as an easy interface to data mining algorithms.
50
Pradhan, Shameer Kumar. "Investigation of Event-Prediction in Time-Series Data : How to organize and process time-series data for event prediction?" Thesis, Högskolan Kristianstad, Fakulteten för naturvetenskap, 2019. http://urn.kb.se/resolve?urn=urn:nbn:se:hkr:diva-19416.
Texto completo da fonteResumo:
The thesis determines the type of deep learning algorithms to compare for a particular dataset that contains time-series data. The research method includes study of multiple literatures and conduction of 12 tests. It deals with the organization and processing of the data so as to prepare the data for prediction of an event in the time-series. It also includes the explanation of the algorithms selected. Similarly, it provides a detailed description of the steps taken for classification and prediction of the event. It includes the conduction of multiple tests for varied timeframe in order to compare which algorithm provides better results in different timeframes. The comparison between the selected two deep learning algorithms identified that for shorter timeframes Convolutional Neural Networks performs better and for longer timeframes Recurrent Neural Networks has higher accuracy in the provided dataset. Furthermore, it discusses possible improvements that can be made to the experiments and the research as a whole.