A fair allocation algorithm for predictive police patrolling:

We address the tradeoff of developing good predictive models for police allocation vs. optimally deploying police officers over a city in a way that does not imply an unfair allocation of resources. We modify a fair allocation algorithm to tackle a real world problem: crime in the city of Bogota, Colombia. Our approach allows for more sophisticated prediction models and we ́show that the whole methodology outperforms the current police allocating mechanism in the city. Results show that even with a simple model such as a Kernel Density Estimation of crime, one can have much better predictions than the current police model and, at the same time, mitigate fairness concerns. Although we can not provide general performance guarantees, our results apply to a real life problem and should be seriously considered by policy makers.