Design of an agent-based model to predict crime (WIP).

In modern societies where fighting crime has a long history, establishing effective methods for crime prevention is of high significance. For this purpose, police departments worldwide undertake efforts to analyze past crime data. They aim to detect the most prolific crime areas and predict their development, in order to direct their prevention efforts. In parallel, scholars investigate possibilities to build crime prediction models, by applying various techniques, from simple regression to data mining. Acknowledging the latest advances in this field, which suggest that Agent-Based Modeling (ABM) is a promising method, in this paper we present the design of an ABM capable of predicting where and when future crimes will most probably happen. We extend the previous work by accounting for offender behavior and integrating a realistic representation of the environment. In contrast to existing models, past crime data will be included to achieve automatic calibration. Furthermore, we will assess how crime data can be used to model agent's behavior and we will include environmental data stepwise, in order to achieve the optimal balance between prediction accuracy and complexity. The resulting ABM will be developed as a crime prediction tool, and as an experimental environment to test prevention strategies.