Unsupervised Segmentation of Time Series Based on Max Entropy Voting Model
Gao Cheng-kai · Jisuanji gongcheng · 2009
Through the high-dimension segmentation, the high-level symbol expression can be created. This paper proposes an unsupervised segmentation algorithm for high-dimension time series. This method can solve the pretreatment problem of high-dimension symbolization. It realizes the clustering of high-dimension data, and uses max entropy voting model to do series segmentation. Experimental results show that the algorithm's average recall ratio and precision ration are respectively 0.86 and 0.88. Its whole performance is better than Principal Component Analysis(PCA) algorithm and Probability Principal Component Analysis(PPCA) algorithm.