spatial-location/time-series prediction models How efficient it is to build predictive model. However, every crime is dependent on three factors: Time, Spatial Location and people behavior. 
Statistically, we can't measure people behavior (we can't use these private data). Hard to find good and accurate results using spatial-location/time-series predictive systems. 
Statistically, Is it valid to use only one dimension (spatial-location or time-series) to analyze and predict events (like crimes) ?
 A: I would use the phrase don't let the perfect be the enemy of the good in this situation. Sure if we had very detailed estimates of the number of people walking around at particular locations we could make better forecasts of crime, but gaining such estimates over a large area of interest (often an entire city for a police department) are infeasible in most situations.
In terms of forecasting, people tend to find the accuracy and variance of forecasts are impacted by the baseline rates of which crime occurs. That is, for more rare crimes it becomes more difficult to forecast, and everything becomes more rare when you shrink either the time or spatial window in which you are trying to forecast. 
I would suggest the following citations, and you can judge for yourself whether the forecast error is reasonable or efficient given your own standards.


*

*Johnson, S. D., Bowers, K. J., and Pease, K. (2012). Towards the modest predictability of daily burglary counts. Policing, 6(2):167-176.

*Cohen, J., Gorr, W. L., and Olligschlaeger, A. M. (2007). Leading indicators and spatial interactions: A crime-forecasting model for proactive police deployment. Geographical Analysis, 39(1):105-127.


I know people are selling software for a lot of money to police departments to make similar forecasts, so I know they think it is reasonable!
