Empirical Comparison of Clustering Methods for Long Time-Series Databases

摘要

This paper presents a comparative study of methods for clustering long-term temporal data. We split a clustering procedure into two processes: similarity computation and grouping. As similarity computation methods, we employed dynamic time warping (DTW) and multiscale matching. As grouping methods, we employed conventional agglomerative hierarchical clustering (AHC) and rough sets-based clustering (RC). Using various combinations of these methods, we performed clustering experiments of the hepatitis data set and evaluated validity of the results. The results suggested that (1) complete-linkage (CL) criterion outperformed average-linkage (AL) criterion in terms of the interpret-ability of a dendrogram and clustering results, (2) combination of DTW and CL-AHC constantly produced interpretable results, (3) combination of DTW and RC would be used to find the core sequences of the clusters, (4) multiscale matching may suffer from the treatment of 'no-match' pairs, however, the problem may be eluded by using RC as a subsequent grouping method.