DDTM: A Distance-Based Data Transformation Method for Time Series Classification

Time series classification (TSC) relies primarily on similarity or dissimilarity measurements. In numerous scenarios, classification accuracy can be improved by removing interference variables. DDTM is proposed, which investigates the changes in the distance between the category and the subsequence on time series to determine if the subsequence influences the determination of the category. The main idea consists of three steps. First, propose a window division strategy to compare the Dynamic Time Warping (DTW) distances between time series categories at different window positions. Next, capture the influence of class division by calculating the average distance under various windows and measure the efficiency of window position for class division using information gain. Finally, transform the series according to the information gain. The research shows that the proposed DDTM method achieves superior classification results.