Machine learning and artificial intelligence can be used to predict crime in advance. Governments often use these tools for predictive policing to deter crime. 


However, early efforts at crime prediction have been controversial. This is because systemic biases in police enforcement are not accounted for.


Researchers at the University of Chicago have developed a new algorithm that forecasts crime by learning that forecasts future crimes one week in advance with about 90 per cent accuracy. The algorithm learns patterns in time and geographic locations from public data on violent and property crimes.


The researchers used a separate model to study the police response to crime by analysing the number of arrests following incidents and comparing those rates among neighbourhoods with different socioeconomic status. 


The study describing the results was recently published in the journal Nature Human Behaviour. 


Crime In Wealthier Areas Resulted In More Arrests


According to the study, the model showed that crime in wealthier areas resulted in more arrests, while arrests in disadvantaged neighbourhoods dropped. However, the model demonstrated that crime in poor neighbourhoods did not lead to more arrests, suggesting bias in police response and enforcement.


In a statement released by University of Chicago, Ishanu Chattopadhyay, senior author on the new paper, said what the researchers saw is that when the system is stressed, it requires more resources to arrest more people in response to crime in a wealthy area. Also, stress on the system draws police resources away from lower socioeconomic status areas.


Two Broad Categories Of Crimes Considered


The study said that the tool was tested and validated using historical data from the City of Chicago around two broad categories of reported events. These events are violent crimes and property crimes. Violent crimes include homicides, assaults, and batteries. Property crimes include burglaries, thefts, and motor vehicle thefts. 


The researchers used historical crime data from Chicago because they were most likely reported to police in urban areas where there is historical distrust and lack of cooperation with law enforcement. These crimes are less prone to enforcement bias, similar to drug crimes, traffic stops and other misdemeanour infractions. 


Earlier, crime prediction techniques used an epidemic or seismic approach, in which crime is depicted as emerging in "hotspots" that spread to surrounding areas. The drawback associated with these tools is that they miss out on the complex social environment of cities. Moreover, they do not consider the relationship between crime and the effects of police enforcement. 


How The New Model Is Unique


James Evans, a co-author on the paper, said spatial models ignore the natural topology of the city. He added that transportation networks respect streets, walkways, train and bus lines, while communication networks respect areas of similar socioeconomic background. The model developed by the researchers at the University of Chicago enables discovery of those connections, he said.


New Model Divides The City Into Spatial Areas 1,000 Feet Across


According to the study, the new model isolates crime by looking at the time and spatial coordinates of discrete events and detecting patterns to predict future events. Instead of relying on traditional neighbourhood or political boundaries, which are subject to bias, the new model divides the city into spatial tiles roughly 1,000 feet across and predicts from within these areas. The model not only performed well with the data from Chicago, but also data from seven other cities in the United States, namely Atlanta, Austin, Detroit, Los Angeles, Philadelphia, Portland, and San Francisco.


Evans said the researchers demonstrated the importance of discovering city-specific patterns for the prediction of reported crime. This generates a fresh view on neighbourhoods in the city, allows researchers to ask novel questions, and lets them evaluate police action in new ways, he explained.


The Tool Should Not Be Used To Direct Law Enforcement


According to the authors, the tool's accuracy does not mean that it should be used to direct law enforcement, with police departments using it to swarm neighbourhoods proactively to prevent crime. The tool should instead be used as a toolbox of urban policies and policing strategies to address crime. 


Chattopadhyay said they created a digital twin of urban environments. If one feeds it with data from what happened in the past, it will tell them what is going to happen in the future. 


However, the model is not devoid of limitations. One can use the 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 one applies all these different variables, they can see how the system evolves in response.