Modeling and Anomalous Cluster Detection for Point Processes Using Process Convolutions

We present a model using process convolutions, which describes spatial and temporal variations of the intensity of events that occur at random geographical locations. An inhomogeneous Poisson process is used to model the intensity over a spatial region with multiplicative spatial and temporal covariate effects. Temporal variation in the structure of the intensity is obtained by employing a time-varying process for the convolution. Use of a compactly supported kernel in the convolution improves the computational efficiency. Additionally, anomalous cluster detection in the event rates is developed based on exceedance probabilities. The methods are demonstrated on data of major crimes in Cincinnati during 2006. Supplementary materials for this article are available online.