WMCA: A Weighted Matrix Coverage Based Approach to Cluster Multivariate Time Series

The variables of multivariate time series (MTS) can be numeric or categorical attribute, but many researches payed attention to numeric attribute. This paper focuses on MTS with mixed attributes. A novel approach of weighted matrix coverage is proposed to judge the neighborhood between MTS based on Singular Value Decomposition (SVD) and a notion about the number of common neighbors (NCN) is introduced to measure the similarities. In turn, a modified hierarchical clustering algorithm is put forward. The experimental results show that our algorithm performs better than the standard hierarchical clustering algorithm based on Dynamic Time Wrapping (DTW) distance metric.