Spatio-Temporal Change Detection Using Granger Causal Relation

抄録

We proposed a method to detect a change in causal relations over a multi-dimensional sequence of events. The method makes use of the proposed modified cluster sequence mining algorithm to extract causal relations in the form of cluster sequence patterns: a pair of clusters of event that has their occurrence time determined significant by Granger causality. We proposed a pattern time signature, a cumulative incidence function of the cluster sequence occurring at any given time. The pattern time signature allows us to infer the appearance and disappearance time of each cluster sequence pattern. We validated our method using synthetic data. The result shows that our algorithm can correctly identify the change in causal relation even under noisy data.