Trended DTW Based on Piecewise Linear Approximation for Time Series Mining

Similarity measure is one of the most important aspects for achieving effectiveness in time series analysis and a variety of similarity methodologies have been proposed. A comprehensive comparison of these methods has been made and it is claimed that the classic method Dynamic Time Warping (DTW) is still very competitive [1]. In this paper we make an improvement to DTW regarding to its weakness: instead of carrying out dynamic programming based on isolated points, we proposed a new dynamic programming procedure based on line segments, which naturally allows the trend information into spatiotemporal data mining and plays a role in the alignment. Experiments show that this method is promising in both improving the effectiveness of similarity measure and speeding up the computation.