Cross-correlation based clustering and dimension reduction of multivariate time series

Abstract

In this paper, we investigate dimension reduction possibilities of multidimensional time series data and we introduce a graph based clustering approach using the cross-correlation between time series. The proposed solution consists of two main steps: introducing a novel similarity measure for measuring cross-correlations and a graph-based clustering technique. These two parts are both compared to existing techniques, including noise tolerance and our solution performs better in a noisy environment. The proposed solution is applied to performance metrics of a specific data processing system in order to identify and efficiently visualize connections among the collected metrics. The introduced method provides a more balanced clustering than classic ones, and it is suitable to reveal dependencies and connections among performance metrics time series data.