Crime rate prediction in the urban environment using social factors

摘要

The aim of this study is to compare different approaches to the problem of forecasting the number of crimes in different areas of the city. During this research we studied three types of predictive models: linear regression, logistic regression and gradient boosting. The predictive factors used in these models have been selected using the feature selection techniques. This approach allowed us to increase the accuracy of predictions and to avoid the model's overfitting. The obtained models were tested on criminal data of the city of Saint-Petersburg. We compared the results of model predictions and determined that gradient boosting is the most appropriate method for the problem of crime rate prediction in certain urban area.