A new segmented time warping distance for data mining in time series database

Comparison of time series is a key issue in data mining of time series database. Variation or extension of Euclidean distance is generally used. However Euclidean distance will vary much when time series is to be stretched or compressed along the time-axis. Dynamic time warping distance has been proposed to deal with this case, but its expensive computation limits its application. In this paper, a novel distance based on a new linear segmentation method of time series is proposed to avoid such drawbacks. Experiment results in this paper show that the proposed method achieves significant speedup up to about 20 times than Dynamic time warping distance without accuracy decrease.