Time Series Classification Based on Dictionary Learning and Sparse Representation

The high-dimensionality, shifting and distortion are considered as the main difficulties in time series classification. In this paper, a novel technique based on dictionary learning and sparse representation is developed to conquer these difficulties. Dictionary, which is a set of series or segments, is able to reconstruct signals sparsely. We use dictionary learning to reconstruct time series and introduce k-nearest neighbor classifier based on dynamic time warping to improve the quality of approximation of dictionary. Sparse coding is used to reconstruct an unseen series. The class label of an unseen series is determined by the reconstructive error. Extensive experiments were conducted to test the effectiveness of the proposed algorithm. Experimental results show that the proposed technique achieved significant improvements compared with some existing algorithms.