Prediction of Crime Hotspots based on Spatial Factors of Random Forest

Abstract

Crime has always been one of the important social issues that people care about. In the problem of urban security, preventing and reducing crime is one of the primary tasks of the police. Crime hotspot prediction can use historical crime data to infer geographic areas where crime may occur in the future. Machine learning is the mainstream method of current crime prediction method.But in the era of big data, more and more data information appears in the eyes of people, it is far from enough to use historical crime data to infer crime hotspots. Therefore, this paper is based on the random forest algorithm, first of all,divides the study areas into four categories according to the hot spot distribution based on the historical crime data: frequent hot areas, common hot areas, occasional hot areas and non-hot areas,and then, representative covariates from the non-historical crime data are added to the prediction model to explore the changes in the result accuracy of crime prediction based on the historical crime data by integrating different covariates. The data is based on real data, and the experimental results show that compared with the inference method based only on historical crime data, the accuracy of the model with covariates is improved compared with that without covariates.