Adaptive Matrix Model for a Crime Forecasting Task

Abstract

In the police activities of any country, there are two major directions: identifying signs of crime preparation and preventing its commission (prediction), as well as crime prevention by eliminating the conditions for its commission (prevention). At the same time, various theories are explaining that the place and time of the crime occur at random, and when certain conditions are met for its commission, which depend on the type of crime and various factors of the objective side of the crime (place, time, mechanism of commission) and object of encroachment. Currently, a huge number of criminal events has been accumulated in the databases of law enforcement agencies over a long time (more than 20 years). These events occurred at a specific place (geolocation) and at a specific date and time. Considering a certain residential area as a closed system with processes taking place according to some laws, it can be assumed that criminal events also occur according to some hidden patterns. In modern science, some technologies enable identifying such hidden patterns in large data arrays – Data Mining. Identification of hidden patterns allows performing the function of predicting the commission of new criminal events in space and time. This article considers one of the technology approaches – Adaptive Matrix Model. As a result, we obtained a fairly simple and small model for crime prediction that produced promising experimental results. Ways to further improve accuracy are suggested.