Extracting texture feature for time series classification

Abstract

Time series exists in many pattern recognition and prediction application in many different industrial fields, such as medicine, biology, economy and others. In this kind of a data analytical tasks, the classification phase is one of the most important phases as it allows us to assign a class to a previously unseen record as precise as possible. In classification, past researches have shown that rules such as the 1-Nearest Neighbour with a distance measure in time domain performs well in a wide variety of application domains. However, there are many time series that are not obvious in time domain. For instance, the classification of chainsaws where the feature that represents this time series would be frequency instead of time. For such classification, an alternative representation would be necessary. In this work, we will investigate the use of images for time series classification. In particularly, we extract texture features from recurrence plots as their graphical nature exposes a structural pattern in the data.