Crime Forecasting Using Spatio-temporal Pattern with Ensemble Learning

Abstract

Crime forecasting is notoriously difficult. A crime incident is a multi-dimensional complex phenomenon that is closely associated with temporal, spatial, societal, and ecological factors. In an attempt to utilize all these factors in crime pattern formulation, we propose a new feature construction and feature selection framework for crime forecasting. A new concept of multi-dimensional feature denoted as spatio-temporal pattern, is constructed from local crime cluster distributions in different time periods at different granularity levels. We design and develop the Cluster-Confidence-Rate-Boosting (CCRBoost) algorithm to efficiently select relevant local spatio-temporal patterns to construct a global crime pattern from a training set. This global crime pattern is then used for future crime prediction. Using data from January 2006 to December 2009 from a police department in a northeastern city in the US, we evaluate the proposed framework on residential burglary prediction. The results show that the proposed CCRBoost algorithm has achieved about 80% on accuracy in predicting residential burglary using the grid cell of 800-meter by 800-meter in size as one single location.