Interpreting a Conditional Generative Adversarial Network Model for Crime Prediction

Crime prediction models seek to assist policymakers and law enforcement agencies in the allocation of scarce resources intended to prevent crime occurrences. This paper proposes an extension and two interpretation methods for a novel conditional GANs architecture for crime (robberies) prediction in Bogota, Colombia. The model’s performance on the area under the Hit Rate - Percentage Area Covered by Hotspots curve increases from 0.86 to 0.88 AUC when extended by conditioning on holidays. The proposed interpretability methods can help study the effect of crime occurring in a region on the likelihood of occurrence in other regions through the use of SHAP values. These interpretations can prove to be very useful for policymakers and law enforcement agencies in designing interventions and preventing future crime from inferring potential regions of displacement. (Results of the project “Diseño y validación de modelos de analítica predictiva para la toma de decisiones en Bogotá” funded by Colciencias with resources from the Sistema General de Regalías, BPIN 2016000100036. The opinions expressed are solely those of the authors .)