Time works well: Dynamic time warping based on time weighting for time series data mining

Abstract

Dynamic time warping is one of the most important similarity measurement methods for time series data mining. Owing to the different influence of various time points, an extension of dynamic time warping based on time weight analysis is proposed, where the weights of pairs of time points from two series can be automatically calculated through measuring how far the history time points are from the latest ones. The time weights of the matching pairs in the warping path obtained by dynamic time warping represent the importance of the corresponding time points and will make different contributions to the accumulated cost matrix. The hierarchical clustering results in various types of time series data, including UCI data and financial stock exchange data, demonstrate that time works wonders, and different history time points have different influence on the contribution of the minimal distance between two time series. Compared to state-of-the-art methods, the proposed technique takes the time factor into consideration and can be advantageously used for similarity measurement in time series data mining.