It hard to argue for any reliable indication of accuracy. What does "less than half a percent" success rate even mean? What if you're sampling an interval where there is no crime? How would you know that the model "failed"? How does "less than half a percent" compare to other means of prediction — like following the gut feeling of an expert?
Well, it would increase detection. Increased detection does not always mean increased crime levels. It could lead to a feedback cycle. More crime (even rather small common crime like j-walking) goes up, increases statistics, causes more funding..
> Well, it would increase detection. Increased detection does not always mean increased crime levels.
surely this is figured out and accounted for. The healthcare industry has been coming up with better tests forever that result in more disease detections. But they don't scream to the hills of a skyrocketing outbreak because they account for the better test.
Beyond the fact that reported crimes are just a proxy for actual crime, this is clearly a complex, non-linear system with feedback loops. I don't see how simple statistics, markov reasoning, ML or AI could ever realistically model this with the intent being control of the surveyed system.
Huge paradox. Why even try predictive policing without Minority Report oracles to magically do the prediction?