Multiple Shape-based Template Matching for time series data

1-Nearest Neighbor classification with Dynamic Time Warping distance measure is mainly used for time series classification. In large datasets, major concerns for the classification problem are CPU time and storage requirement. Recently, Shape-based Template Matching Framework (STMF) was proposed to resolve these problems by constructing a template as a representative for each class of the data, and then STMF uses these templates to classify a query sequence. However, a single template per class may not well represent the overall characteristic of the data. In this paper, we propose a new method called Multiple Shape-based Template Matching (MSTM) extended from STMF. Our method constructs multiple templates by clustering each class of data and also learning the global constraint to increase the accuracy. In the experiment, we evaluate by comparing with STMF which uses only one template per class and the original 1-NN classification with global constraint. Our proposed method also minimizes the number of templates and still classifies the query sequence effectively.