LB HUST: A Symmetrical Boundary Distance for Clustering Time Series

Clustering is an important technology in mining time series, and the key is to define the similarity or dissimilarity between data. One of existing time series distance measures LB_Keogh, is tighter lower bounding than Euclidean and dynamic time warping (DTW), however, it is an asymmetrical distance measure, and has its limitation in clustering.To solve the problem, we present a symmetrical boundary distance measure called LB_HUST, and prove that it is tighter lower bounding than LB_Keogh. We apply LB_HUST to cluster time series, and update the boundary of the cluster when a new time series is added into the cluster. The experiments show that the method exceeds the approaches based on Euclidean and DTW in terms of accuracy.