Automated Change Detection and Reactive Clustering in Multivariate Streaming Data

Many automated systems need the capability of automatic change detection without the given detection threshold. This paper presents an automated change detection algorithm in streaming multivariate data. Two overlapping windows are used to quantify the changes. While a window is used as the reference window from which the clustering is created, the other called the current window captures the newly incoming data points. A newly incoming data point can be considered a change point if it is not a member of any cluster. As our clustering-based change detector does not require detection threshold, it is an automated detector. Based on this change detector, we propose a reactive clustering algorithm for streaming data. Our empirical results show that our clustering-based change detector works well with multivariate streaming data. The detection accuracy depends on the number of clusters in the reference window, and on the window width.