Hierarchical Clustering of Time Series Based on Linear Information Granules

Time series clustering is one of the main tasks in time series data mining. In this paper, a new time series clustering algorithm is proposed based on linear information granules. First, we improve the identification method of fluctuation points using threshold set, which represents the main trend information of the original time series. Then using fluctuation points as segmented nodes, we segment the original time series into several information granules, and linear function is used to represent the information granules. With information granulation, a granular time series consisting of several linear information granules replaces the original time series. In order to cluster time series, we then propose a linear information granules based segmented matching distance measurement (LIG_SMD) to calculate the distance between every two granular time series. In addition, hierarchical clustering method is applied based on the new distance (LIG_SMD_HC) to get clustering results. Finally, some public and real datasets about time series are experimented to examine the effectiveness of the proposed algorithm. Specifically, Euclidean distance based hierarchical clustering (ED_HC) and Dynamic Time Warping distance based hierarchical clustering (DTW_HC) are used as the compared algorithms. Our results show that LIG_SMD_HC is better than ED_HC and DTW_HC in terms of F-Measure and Accuracy.