Ok, so first, if the crime doesn't happen, how do you know you prevented it? Maybe it just didn't happen.
You don't look at individual crimes, you take a selection of areas with similar crime statistics, implement the prediction system in some of them, then see how the crime rates change.
Second, doesn't this seem like there will now be a market for anti-prediction? That is, find out where the cops think the crime will occur, and do the crime somewhere else. Because the cops will be somewhere else, your chances of getting caught are less
Perhaps there will be a market for anti-prediction, but the types of crimes that this aims to prevent (or even just be more response to calls about) aren't usually done by sophisticated criminals. Any anti-prediction system would first have to be able to aggregate crime statistics then apply the prediction algorithm, then find areas outside the predicted zones. If you have all that already, you might as well just sell the prediction algorithm to the police rather than make an unethical program that has a very small (and secretive) user base that wouldn't pay much for your system in the first place.