Time series are very common in presenting collected data such as economic indicators, natural phenomenon, control engineering data, among others. In the last decade, the interest in time series data mining increased as the amount of collected data increased dramatically. Standard approaches for time series classification are based on collecting distance measures, such as the Euclidian distance (ED) and dynamic time warping (DTW) along with 1-NN classifier for further classification. Recently, more advanced types of classification were found, introducing primitives (named time series shapelet) that consistently represent a certain class. The time series shapelet is a small sub-section of the entire time series, which is “particularly discriminating”. It appears that shapelets based classification produces higher accuracies on some data sets, based on the fact that the global features are more sensitive to noise than locals. Despite its advantages, the time series shapelets classification has an apparent disadvantage: very slow training time. This work attempts to improve the training time for the originally proposed time series shapelets classification algorithm and introduces a new approach for time series classification based on concatenated decision tree paths. First, the classical algorithm for time series classification based on shapelets, is significantly improved in terms of the training time. The improvement is based on using randomly generated sequences tuned in a particle-swarm-optimization (PSO) environment, instead of using sub-series from the original time series. Second, a new highly accurate classification method, based on concatenated decision tree paths, is introduced. The approach builds a unique representative pattern of a certain class based on the taken paths in a pool of decision trees. Third, the proposed method has been successfully extended for a 2-class-labels classification problem where only one decision tree can be built. A variety of 2-class-labels decision trees were built based on different splitting criterion (distance to a random shapelet); thus- increasing the pool of decision trees and increasing the overall accuracy. Fourth, the proposed method was successfully applied on two classes image classification problem, by converting the image into time series. An accuracy of around 95% was achieved for the pedestrian detection case from the Daimler database.

Sensor-based human activity recognition has become an important research field within pervasive and ubiquitous computing. Techniques for recognizing atomic activities such as gestures or actions are mature for now, but complex activity recognition still remains a challenging issue. I was a candidate in an activity classification thesis. It collected 4 activities, which included walking on the sidewalk for a set distance, walking up and down a set of stairs, walking on the treadmill at 2.5 mph for 2 minutes, and jogging on the treadmill at 5.5 mph for 1 minute. It took 30 minutes to collect one candidate data. If complex activity data can be made up with atomic activities data, the data collecting process will be simplified. In this thesis, I used methods to mimic a complex activity shapelet by combing atomic activity shapelets. I first collect two candidates walk, jump and skip time series data, in which walk and jump are considered the atomic activities of skip. Time series patterns, shapelets, are extracted using tsshapelet package. Shapelets are small sub-series, or parts of the time-series, that are informative or discriminative for a certain class. They can be used to transform the time-series to features by calculating the distance for each of the time-series you want to classify to a shapelet. In order to create skip representative shapelet, Barycenter Dynamic Time Warping and Weighted Dynamic Time Warping are used to average walk and jump shapelet, and then compare the euclidean distance between skip shapelet with walk shapelet, jump shapelet and, combined-shapelet. Experimental result show that the combined-shapelet is closer to skip shapelet than single walk or jump shapelet. Then I use three evaluation methods to mathematically and statistically show that combined-shapelet and real skip shapelet are similar. Evaluation methods include sliding window, cycle comparison and random comparison. To verify whether combined-shapelet can substitute real skip shapelet, a new labeled time series data is introduced, the result shows that both shapelets have the label accuracy around 70%, accuracy difference is less than 1%.

[pt] Identificar o modelo de um reservatório é o primeiro passo para interpretar corretamente os dados gerados em um teste de poços e desta forma estimar os parâmetros relacionados a esse modelo. O objetivo deste trabalho é de forma inversa, utilizar as curvas de pressão obtidas em um teste de poços, para identificar o modelo de um reservatório. Como os dados obtidos em um teste de poços podem ser ordenados ao longo do tempo, nossa abordagem será reduzir essa tarefa a um problema de classificação de séries temporais, onde cada modelo de reservatório representa uma classe. Para tanto, foi utilizada uma técnica chamada shapelet, que são subsequências de uma série temporal que representam uma classe. A partir disso, foi construído um novo feature space, onde foi medida a distância entre cada série temporal e as shapelets de cada classe. Então foi criado um comitê de votação utilizando os modelos k-nearest neighbors, decision tree, random forest, support vector machines, perceptron, multi layer perceptron e adaboost. Foram testados os pré-processamentos standard scaler, normalizer, robust scaler, power transformer and quantile transformer. Então a classificação foi feita no novo feature space pré-processado. Geramos 10 modelos de reservatório multiclass analíticos para validação. Os resultados revelam que o uso de modelos clássicos de aprendizado de máquina com shapelets, usando os pré-processamentos normalizer e quantile trasformer alcança resultados sólidos na identificação dos modelos de reservatório. / [en] Identifying a reservoir model is the first step to correctly interpret the data generated in a well-test and hence to estimate the related parameters to this model. The goal of this work is inversely to use the pressure curves, obtained in a well-test, to identify a reservoir model. Since the data obtained in a well-test can be ordered over time, we reduce this task to a problem of time series classification, where every reservoir model represents a class. For that purpose, we used a technique called shapelets, which are times series subsequences that represent a class. From that, a new feature space was built, where we measured the distance between every time series and the shapelets of every class. Then we created an ensemble using the models k-nearest neighbors, decision tree, random forest, support vector machines, perceptron, multi-layer perceptron, and adaboost. The preprocessings standard scaler, normalizer, robust scaler, power transformer, and quantile transformer were tested. Then the classification was performed on the new preprocessed feature space. We generated 10 analytical multiclass reservoir models for validation. The results reveal that the use of classical machine learning models with shapelets, using the normalizer and quantile transformer preprocessing, reaches solid results on the identification of reservoir models.

No description available.

La classification de séries temporelles a suscité beaucoup d’intérêt au cours des dernières années en raison de ces nombreuses applications. Nous commençons par proposer la méthode Dense Bag-of-Temporal-SIFT-Words (D-BoTSW) qui utilise des descripteurs locaux basés sur la méthode SIFT, adaptés pour les données en une dimension et extraits à intervalles réguliers. Des expériences approfondies montrent que notre méthode D-BoTSW surpassent de façon significative presque tous les classificateurs de référence comparés. Ensuite, nous proposons un nouvel algorithmebasé sur l’algorithme Learning Time Series Shapelets (LTS) que nous appelons Adversarially- Built Shapelets (ABS). Cette méthode est basée sur l’introduction d’exemples adversaires dans le processus d’apprentissage de LTS et elle permet de générer des shapelets plus robustes. Des expériences montrent une amélioration significative de la performance entre l’algorithme de base et notre proposition. En raison du manque de jeux de données labelisés, formatés et disponibles enligne, nous utilisons deux jeux de données appelés TiSeLaC et Brazilian-Amazon. / Time Series Classification (TSC) has received an important amount of interest over the past years due to many real-life applications. In this PhD, we create new algorithms for TSC, with a particular emphasis on Remote Sensing (RS) time series data. We first propose the Dense Bag-of-Temporal-SIFT-Words (D-BoTSW) method that uses dense local features based on SIFT features for 1D data. Extensive experiments exhibit that D-BoTSW significantly outperforms nearly all compared standalone baseline classifiers. Then, we propose an enhancement of the Learning Time Series Shapelets (LTS) algorithm called Adversarially-Built Shapelets (ABS) based on the introduction of adversarial time series during the learning process. Adversarial time series provide an additional regularization benefit for the shapelets and experiments show a performance improvementbetween the baseline and our proposed framework. Due to the lack of available RS time series datasets,we also present and experiment on two remote sensing time series datasets called TiSeLaCand Brazilian-Amazon