Crime prediction with graph neural networks and multivariate normal distributions

We study high-resolution crime prediction and introduce a new generative model applicable to any spatiotemporal data with graph convolutional gated recurrent units (Graph-ConvGRU) and multivariate Gaussian distributions. We introduce a subdivision algorithm and create a graph representation to tackle the sparsity and complexity problem in high-resolution spatiotemporal data. By leveraging the flexible structure of graph representation, we model the spatial, temporal, and categorical relations of crime events and produce state vectors for each region. We create a multivariate probability distribution from the state vectors and train the distributions by minimizing the KL divergence between the generated and the actual distribution of the crime events. After creating the distributions, crime can be predicted in any resolution as the first time in the literature. In our experiments on real-life and synthetic datasets, our model obtains the best score with respect to the state-of-the-art models with statistically significant improvements. Hence, our model is not only generative but also precise. We also provide the source code of our algorithm for reproducibility.