Mixed spatial-temporal characteristics based Crime Hot Spots Prediction

Crime Hot Spots refer to the areas in which the crime rates are above the average level, therefore the Hot Spots Prediction is the primary mission of the Public Security Prevention and Control. By encoding the area-specific crime incidents, the crime hot spots has been classified them into different heat levels, rendering the conversion of Hot Spots prediction into a multi-class classification problem. The new prediction model uses time sequence of area-specific heat levels, temporal distance of important holidays, and neighborhood features to establish the crude mixed spatial-temporal characteristics. As with rotational invariance, we use histogram-based statistical methods to design neighborhood features of heat levels. Finally LDA (Linear Discriminant Analysis) is adopted for dimensionality reduction of mixed spatial-temporal characteristics, and KNN is adopted for prediction. Experimental results show that when crime statistics are conducted on a “Weekly” basis, the new prediction model can achieve optimal performance.