Corso di laurea magistrale Interateneo in Fisica dei sistemi complessi

Time series classification is the task of assigning a label to the given time series. This has applications in several fields, such as finance, medicine and industry. To solve this task, in the literature, Shapelets have been proposed [1]. Shapelets are subsequences of a time series that capture the characteristics of a certain class of time series in a dataset. Using these Shapelets, it is possible to cluster the time series in the dataset, by assigning a time series the label of a class based on the presence or absence of the candidate Shapelet representative of that class [2]. Shapelets can be used in the context of unsupervised anomaly detection [3], allowing to identify the anomalous time series based on the absence of normality characteristics learned by the model through the candidateShaplets ofthe normal class.

The goal of this thesis is to study the recent advances in the classification of time series using Shapelets, with particular focus on unsupervised models oriented towards anomaly detection and to apply these algorithms to a real industrial problem.