Predicting the Risk of Bicycle Theft Occurrence Considering Routine Activity Theory and Spatial Correlation

The purpose of this study is to predict the risk of bicycle theft with high accuracy. We hypothesized that a model that takes into account spatial correlations would improve prediction accuracy by allowing the model to reflect near repeat victimization, one of the phenomena of crime. We employed the Poisson CAR (conditional auto-regressive) model as a model that takes spatial correlation into account and conducted forecasting. For performance comparison, we also used the Poisson regression model, which does not take spatial correlation into account, to make predictions. As a result, the Poisson CAR model had higher prediction accuracy than the ordinary Poisson regression model.