Abstract

We propose a number of alternative modelling frameworks for spatial and spatio-temporal point patterns living on networks. We first consider spatio-temporal Dirichlet process mixture models on linear network windows to automatically detect space-time clusters of the events. A Bayesian hierarchical model is adopted with a convolution kernel estimator to account for the network structure in the city. We then use marked point patterns, where the complexity of the marks is
accounted through GAT neural networks. Gibbs point processes are then developed for network data.