Identifying Propagating Signals with Spatio-Temporal Clustering in Multivariate Time Series

Jan David Hüwel, Georg Stefan Schlake, Kevin Albrechts, Christian Beecks

Abstract: Recordings via multiple sensors positioned on a grid result in multivariate time series, whose subsequences can be compared in aspects of time, space and shape. Existing methods for tracking propagating signals in this high-content data format require a starting position, which has to be determined by a domain expert. In this paper, we propose a fully unsupervised method to discover propagating signals via density-based clusters with respect to the three aspects mentioned above. For this purpose, we adapt the DBSCAN algorithm to our specific setting and present an exemplary application of this method on pharmacological data.