Early classification of spatio-temporal events using time-varying models

Sevvandi Kandanaarachchi, Rob J Hyndman and Kate Smith‑Miles


This paper investigates early event classification in spatio-temporal data streams. We propose a framework for early classification that considers the relationship between the features of an event and its age. The framework incorporates an event extraction algorithm as well as two early event classification algorithms, which use a series of logistic regression classifiers with penalty terms and state space models. We apply this framework to synthetic and real world problems and demonstrate its reliability and broad applicability. The algorithms are available in the R package eventstream, and other code in the supplementary material.