Efficient Time Series Classification under Template Matching Using Time Warping Alignment

One of the most widely used time series classification is the 1-nearest neighbor (1-NN) classification algorithm which utilizes dynamic time warping (DTW) as a similarity measure. On large training data, though DTW is demonstrated to be highly accurate, its 1-NN classification typically takes significant amount of time to classify a given test sequence. The hotspot for this type of computation lies in the repeated DTW computations. In limited storage applications such as some real-time embedded systems, there might not be sufficient amount of resources for such computation. In this paper, we propose a novel template construction algorithm based on the accurate shape averaging (ASA) technique. Each training class is represented simply by only one sequence. Our experiments show that the 1-NN classification with our proposed template construction algorithm can gain significant performance improvement while maintaining its high accuracy.