Extracting meaningful patterns for time series classification
Xiaohang Zhang, Jun Hui Wu, Xuecheng Yang, Ting Lu · 2008
An import area in machine learning is multivariate time series classification. In this paper we present a novel algorithm which extracts some meaningful patterns from time series data and then uses traditional machine learning algorithm to create classifier. During the stage of pattern extraction, the Gird function is used to evaluate the patterns and the starting position and the length of each pattern are automatically determined. We also apply sampling method to reduce the search space and improve efficiency. The common datasets are used to check our algorithm which is compared with the naive algorithms. The results show that a lot of improvement can be gained in terms of interpretability, simplicity of the model and also in terms of accuracy.