A Heuristic Approach to Crime Prediction based on Generalization of Crime Categories

Background: As crime rates are increasing all over the world, many methods for crime prediction based on data mining have been proposed in the past. Crime prediction finds application in areas like predictive policing, Hotspot evaluation and geographic profiling. It has been observed in the past that crime is closely related to geographical location, time, weather conditions and day of the week. Objective: Thus, to tackle crime events, a proactive policing approach can be developed using crime prediction. The main objective of this study is to provide a heuristic approach to crime prediction. Methods: In this work, a crime prediction approach is proposed which utilizes a crime history dataset which contains multiple categories of crime. And a heuristic approach based on the generalization of crime categories is proposed. A spatiotemporal crime prediction technique based on machine learning techniques is proposed. State-of-the-art classification approaches along with ensemble learning approach are used for prediction. Results: The performance of the proposed model is compared using state-of-the-art classification techniques without a heuristic approach and with a heuristic approach, and it is found that the model with heuristics achieves better accuracy. Conclusion: Crime events dataset can be utilized to predict future crime events in an area because crime shows geographical patterns. These spatial patterns might vary with the category of crime and it is challenging to deal with lots of crime categories. Thus, a generalization based approach can be a vital asset in crime prediction.