A Hybrid Time Series Matching Algorithm Based on Feature-Points and DTW

Feature-points based time series approximation representation utilizes the tendency information, but lacks consideration of the details. DTW (Dynamic Time Warping) based similarity measurement eliminates the time line warp, but computation complexity is high. Based on feature-points and DTW, this paper proposed a hybrid time series matching algorithm. Firstly, we extracted the feature-points of time series as the coarse-grained representation, calculated the DTW distance between feature-points; then applied uniform sampling on feature-points segmentations as the fine-grained representation, calculated the Euclidean distance between corresponding segmentations; at last, we summed the two distances as the final distance. The algorithm achieved a high matching accuracy while lowered the computation overhead. This paper used several time series data sets from UCR to do the experiments, verified the effectiveness of the proposed algorithm.