Guest Lecture Notes: Predictive Policing

By: Aaron Boogaard and Jonne Frankena, students Minor Managing Digital Innovation

After helping the class get over flashbacks of Minority Report and Tom Cruise, Mr. Dick Willems gave an interesting lecture about Predictive Policing for the minor course New Ways of Working. Mr. Willems, who studied mathematical psychology and now works as a data scientist for the police department of Amsterdam, talked about building and implementing an analytical model for predicting crimes. This related very well to the literature discussed in the course, which made the literature much more lively.

Building the Predictive model

In order to predict the area in which a crime might take place, historical data about the crime is combined with recent trends and other data points. Willems calls this the Crime Anticipation System (CAS). The goal of CAS is: “to allocate capacity as efficiently as possible, so that police forces are deployed where it matters, at those times it matters most”. With a logistic regression model, CAS finds some areas where the chance of the next crime is highest, and based on this information an analyst and planner can discuss what to do next. Willems’ discussion of the challenges of building a model for different crimes nicely illustrates how knowledge can be both objective and subjective at the same time. The judgment of how much historical data, trends and which other data points should be used, is to some degree subjective as it is based on the expertise of Mr. Willems. Going back longer in time or using other data points would yield different outcomes. The results of the model, however, will often be interpreted as objective by those who use it.

Implementation of the model

One of the things we found great about Mr. Willems’ approach is that he doesn’t just make the model available for everyone in the organization. Instead, he makes sure that there will be a local analyst that builds on the model. As Willems said the model will only tell when and where a crime might happen, not why or how. The local analysts play an important role in asking the why question and trying to figure out the how. By talking with the police officers on the streets and other experts, they can try to determine the causality and discuss with a planner what would be the most effective measures to prevent the crime. This way, Willems safeguards against one of the often discussed traps of big data: looking for correlation, instead of causation.

While CAS was developed for the Amsterdam Police Department, the new National Police Organization is showing an interest to apply the system in other departments as well. Over time with additional development and resources, it might take more data into account and move from logistic regression to a neural network. In the future, Willems thought it would be interesting for all officers to have access to the planner’s map via smartphone, and to encourage them to patrol the designated areas by the use of gamification.

These are just some topics from the lectures that stood out for us as they so closely align with what we discussed in the lectures. Besides these topics, it was intriguing to hear Mr. Willems talk about the use predictive policing, patrolling and the dynamics of analysts and police officers. While the CAS model is not to the point of clairvoyance like the characters in Minority Report (yet?), it does help the police to fight crime more effectively.

We would like to thank Mr. Willems for his guest lecture and for the work he and the police are doing to prevent crimes.