Performance of Mahalanobis Distance in Time Series Classification Using Shapelets

Abstract

Time series data are sequences of values measured over time. One of the most recent approaches to classification of time series data is to find shapelets within a data set. Time series shapelets are time series subsequences which represent a class. In order to compare two time series sequences, existing work uses Euclidean distance measure. The problem with Euclidean distance is that it requires data to be standardized if scales differ. In this paper, we perform classification of time series data using time series shapelets and used Mahalanobis distance measure. The Mahalanobis distance is a descriptive statistic that provides a relative measure of a data point's distance (residual) from a common point. The Mahalanobis distance is used to identify and gauge similarity of an unknown sample set to a known one. It differs from Euclidean distance in that it takes into account the correlations of the data set and is scale-invariant. We show that use of Mahalanobis distance measure instead of Euclidean distance measure in time series dataset classification using shapelets leads to increase in accuracy.