Random-shapelet: an algorithm for fast shapelet discovery
Résumé
—Time series shapelets proposes an approach to extract subsequences most suitable to discriminate time series belonging to distinct classes. Computational complexity is the major issue with shapelets: the time required to identify interesting subsequences can be intractable for large cases. In fact, it is required to evaluate all the subsequences of all the time series of the training dataset. In the literature, improvements have been proposed to accelerate the process, but few provide a solution that dramatically reduces the time required to find a solution. We propose a random-based approach that reduces the time necessary to find a solution, in our experimentation until 3 orders of magnitude compared to the original method. Based on extensive experimentations on several data sets from the literature, we show that even with a few time available, random-shapelet algorithm is able to find very competitive shapelets. I. INTRODUCTION Handling time series in the field of machine learning is not an easy task, in particular due the high dimensionality of the data and because the information is temporal. Among the approaches that have been proposed in the last decade, time series shapelets [13], [14] is a promising technique, in particular because it gives competitive accuracies for time series classification tasks. A shapelet is a subsequence, extracted from a time series, that is the most representative of a class: its occurrence in a time series is characteristic of a class label. For example, we can think about the patterns in heart-rate time series associated with a particular pattern of regular functioning and another pattern associated with a disease. The shapelet discovery relies on two main steps: • The enumeration of all possible subsequences of a training set of time series. These subsequences are called shapelet candidates. • The evaluation of all the shapelet candidates. Shapelet candidates that best separate the classes on the training set are retained. This approach holds desirable properties such as interpretabil-ity and velocity to produce a classification, once the training phase has been achieved.