Multiscale and Multivariate Time Series Clustering: A New Approach.

This paper introduces a new approach to multiscale and multivariate time series clustering based on the X-MeansTS method. It is common that the notion of multivariate time series clustering is defined as the grouping of a set of time series, and not the grouping of instances described by a set of common time series. For this second case, the proposed method is the first approach that meets the constraints of this type of clustering formalized in this paper. The quality of a clustering depends strongly on the distance measure used. The choice of the measure can also depend on the domain; in some cases, the clustering of some time series is done by considering more the interval of the measures on the y-axis while considering the shift of the series on the x-axis. The proposed multivariate method remains robust to these shifts.