Can AI predict crime rates based on historical data, weather patterns and other sensory data ?

AI can now produce short-term, localized crime-risk forecasts by fusing historical incident data with real-time feeds such as weather, foot-traffic sensors, social-media chatter and even gunshot-detection arrays. Modern systems use spatiotemporal deep-learning models (e.g., graph neural networks over geographic grids and transformer-based sequence learners) that outperform older statistical methods on several municipal datasets, reaching 15–30 % gains in precision-recall metrics for the next-shift hotspot prediction task. These tools are deployed in a handful of U.S. and European cities, primarily for resource-allocation rather than individual-level targeting, and are subject to ongoing evaluation for fairness and bias against underserved neighborhoods. At present, medium-range forecasts (weeks or months ahead) remain far less reliable, and most agencies treat AI outputs as decision-support rather than definitive evidence.

— Enriched May 12, 2026 · Source: National Institute of Justice