Spatio-temporal clustering benchmark for collective animal behavior

Abstract

Various spatio-temporal clustering methods have been proposed to detect groups of jointly moving objects in space and time. However, such spatio-temporal clustering methods are rarely compared against each other to evaluate their performance in discovering moving clusters. Hence, in this work, we present a spatio-temporal clustering benchmark for the field of collective animal behavior. Our reproducible benchmark proposes synthetic datasets with ground truth and scalable implementations of spatio-temporal clustering methods. The benchmark reveals that temporal extensions of standard clustering algorithms are inherently useful for the scalable detection of moving clusters in collective animal behavior.