Mixed spatial-temporal characteristics based Crime Hot Spots Prediction

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