Predictive policing is about planning specific (geographical) policing activities based on predictions created by data analytics, whereas traditional policing operations rely on human risk assessment. Sophisticated algorithms to crunch big data will enable the police to predict crimes. Predictive policing enables the police to take preventive measures to mitigate or reduce crime risks, e.g. by being present at locations where crime probability is the greatest. An increasing number of police departments are now adopting it, although there are still questions about its effectiveness.
This paper focuses on the data and algorithms in the predictive policing feedback loop and shows how the current approaches and algorithms actually have two fundamental flaws, namely the 'reporting bias' and one we named the 'back-to-the-future paradox'. We propose a new approach for the predictive policing feedback loop to counter this flaw and show by means of simulation that it is able to overcome these flaws.