Abstract

Police agencies expend considerable effort to anticipate future incidences of criminal behaviour. Since a large proportion of crimes are committed by a small group of individuals, preventive measures are often targeted on prolific offenders. There is a long-standing expectation that new technologies can improve the accurate identification of crime patterns. Here, we explore big data technology and design a machine learning algorithm for forecasting repeated arrests. The forecasts are based on administrative data provided by the national Chilean police agencies, including a history of arrests in Santiago de Chile and personal metadata such as gender and age. Excellent algorithmic performance was achieved with various supervised machine learning techniques. Still, there are many challenges regarding the design of the mathematical model, and its eventual incorporation into predictive policing will depend upon better insights into the effectiveness and ethics of preemptive strategies.