When evaluating crime prediction over specific crime categories, I tried not to normalize the traffic flow. The observation is surprising. Under most cases, without normalization on taxi flow significantly improve the performance. However, for other crime categories, the relative error exploded (200% - 1100% times over crime).
When evaluating crime prediction over specific crime categories, I tried not to normalize the traffic flow. The observation is surprising. Under most cases, without normalization on taxi flow significantly improve the performance. However, for other crime categories, the relative error exploded (200% - 1100% times over crime).
Results are available here