Similarity Measure of Time Series Based on Feature Extraction

Abstract

Time series data mining attracts a lot of attentions in many applications. Similarity measure of time series is a common problem in data mining tasks. Among the existing similarity measures, dynamic time warping can get high accuracy, but the computational cost is expensive. In this paper, we first segment sequences and extract their features; then, a similarity measure is proposed to balance the contradiction between computational cost and accuracy. Finally, experiments are implemented on time series data sets. The experimental results show that our proposed method can effectively improve the computational efficiency of similarity measure.