FEDERJA

keywords: Agent-Based Modeling and Simulation, Predictive Policing, Data Science, Multi-Agent Reinforcement Learning

In some countries, such as the USA and France, police forces have been using methods for analyzing and predicting crime and other offenses on a regular basis for years. However, the few existing initiatives in this area in Spain are very localized (mainly in Madrid) and difficult to extrapolate to other provinces. Publicly available information, combined with data from local authorities, social services and information gathered by police officers in the field, can help law enforcement agencies and supranational bodies (Europol, Interpol, etc.) to anticipate potential trouble spots. However, despite the availability of this information, police intelligence units are not able to process and analyze all this information in order to gain useful knowledge that would allow them to identify and apply containment strategies in potential trouble spots before crimes occur. For the last year, the main researchers of this project have been working with the Intelligence Unit (UTI) of the National Police of Málaga (CNPM), within the framework of a collaboration agreement. As a first step in this collaboration, the CNPM has provided a set of data on crime in the province of Málaga for study.

The main objective of this project was to provide funding for this collaboration in order to develop an ecosystem of predictive models at two levels (macro and micro) that will enable the LEAs to be more efficient, better plan its resources and, as a result, reduce and even prevent crime. The idea was to identify, on the basis of the data provided by the CNPM, those scenarios, mainly conditioned by two factors: space and time, that can give rise to potential risk situations. Two levels of predictive models were developed: on the one hand, those relating to the behavior patterns of offenders (micro-modeling), which aim to infer relationships between the information available on crimes, suspects and certain events and characteristics (intelligence analysis). On the other hand, those relating to criminal activity (both recent and potential) by analyzing how, when and where in order to identify patterns and trends in crime (tactical analysis). The latter integrated socio-demographic and spatial factors to identify long-term patterns and evaluate law enforcement responses and procedures (strategic analysis) (macro-modeling). To achieve these goals, we used agent-based modeling and simulation combined with data science techniques.

Principal investigators:

• Eduardo Guzmán
• María Victoria Belmonte

Researchers:

• José Luis Pérez de la Cruz
• Eva Millán
• Lawrence Mandow
• Juan Palma Borda
• Sergio Gavilán
• Pablo Ruiz-Cruces