Crime Prediction Algorithm Reveals Police Bias in Response
Governments want to use machine learning and AI for predictive policing to deter crime. Early efforts at crime prediction were controversial because they ignored police biases and the complex relationship between crime and society.
University of Chicago data and social scientists developed a new algorithm that forecasts crime by learning time and geographic patterns from violent and property crimes. The model is 90% accurate one week in advance.
In a separate model, the research team compared the number of arrests after incidents in neighborhoods with different socioeconomic status. Wealthier areas had more crime and more arrests, while poorer areas had fewer. Crime in poor neighborhoods didn’t lead to more arrests, suggesting police bias.
“What we’re seeing is that when you stress the system, it requires more resources to arrest more people in response to crime in a wealthy area and draws police resources away from lower socioeconomic status areas,” said Ishanu Chattopadhyay, Ph.D., Assistant Professor of Medicine at UChicago and senior author of the new study.
The tool was tested and validated using Chicago’s historical data on violent (homicides, assaults, and batteries) and property crimes (burglaries, thefts, and motor vehicle thefts). These crimes were most likely to be reported to police in urban areas with a history of distrust and lack of cooperation. These crimes are less prone to enforcement bias than drug crimes, traffic stops, and other misdemeanors.
Previous crime prediction attempts used an epidemic or seismic approach, depicting crime as emerging in “hotspots” that spread. These tools ignore the social complexity of cities and the relationship between crime and police enforcement.
“Spatial models ignore the natural topology of the city,” says sociologist and co-author James Evans, Ph.D. “Transportation networks respect streets, walkways, train and bus lines. Communication networks respect areas of similar socio-economic background. Our model enables discovery of these connections.”
The new model isolates crime by analyzing the time and space of discrete events to predict future events. It divides the city into 1,000-foot-square tiles to predict crime instead of using biased neighborhood or political boundaries. The model worked for Atlanta, Austin, Detroit, Los Angeles, Philadelphia, Portland, and San Francisco.
“We demonstrate the importance of discovering city-specific patterns for the prediction of reported crime, which generates a fresh view on neighborhoods in the city, allows us to ask novel questions, and lets us evaluate police action in new ways,” Evans said.
Chattopadhyay says the tool’s accuracy shouldn’t be used to direct law enforcement to swarm neighborhoods to prevent crime. Instead, it should be added to a toolbox of urban crime-fighting strategies.
“We created a digital twin of urban environments. If you feed it data from happened in the past, it will tell you what’s going to happen in future. It’s not magical, there are limitations, but we validated it and it works really well,” told Chattopadhyay. “Now you can use this as a simulation tool to see what happens if crime goes up in one area of the city, or there is increased enforcement in another area. If you apply all these different variables, you can see how the systems evolves in response.”
Event-level prediction of urban crime reveals enforcement bias in U.S. cities. Victor Rotaru, Yi Huang, and Timmy Li are also authors.
Information from phys.org was used in this report
Featured Image: Steven Spielberg’s Minority Report (2002).