Crime prediction using patterns and context

Science fiction had anticipated the prediction of future occurrence of crimes. In fact, that prediction is actually possible. It can be done with some imprecision and by computer algorithms using available data from various sources. The prediction involves approximate time and risk maps of occurrence of certain type of felonies such as home burglaries, armed robberies and violent thefts. The police can then use this information for increasing their patrolling accordingly and thereby reducing the crime occurrence rate. We present a crime prediction solution developed for Chilean large cities. Its novel approach includes three independent software modules which make predictions based on different algorithms. The final prediction is the cooperative integration of the individual ones. The developed system has been tested on historical data and its performance has been considered acceptable for police field use. An interesting result is that the performance of each individual module is inferior to the joint performance, validating a hypothesis that different algorithms may exploit different features of the available data.