Many police departments now use AI to predict crime hotspots and allocate patrol resources. The clustered nature of crime data makes Hawkes processes a popular choice. However, missing data due to non-reporting biases predictive models, resulting in inaccurate hotspot forecasts and uneven policing, particularly in vulnerable communities. Our work presents a Wasserstein Generative Adversarial Networks (WGAN) method to address unreported crimes in Spatiotemporal Hawkes models, showing that this approach enhances parametric estimation accuracy despite missing data, leading to more effective policing strategies. This work is available in the preprint at https://arxiv.org/pdf/2502.07111.

Our work includes a case study about the effect of unreported crimes in Bogota that has previously been studied by Akpinar et al. https://arxiv.org/abs/2102.00128, and some of the codes we use are adopted/partially based on the GitHub repo https://github.com/nakpinar/diff-crime-reporting.