Researchers at the University of Chicago have developed an algorithm that predicts crimes a week ahead within a radius of 300 metres with 90% accuracy. The work was published in the journal Nature.
According to the researchers, the system studies patterns drawn from publicly available data on violent offences and property crimes. Such events are less prone to police bias, the researchers say.
The model isolates crime by considering the temporal and spatial coordinates of individual events and identifying patterns to forecast future incidents.
The algorithm divides the city into “spatial tiles” about 300 metres across and predicts offences in those areas.
The researchers said that previous models relied more on traditional neighbourhoods or political boundaries, which are prone to bias.
According to the study’s lead author Ishan Chattopadhyay, the instrument’s high accuracy does not mean it should be used to govern police policy. He added that police administrations should not use it for pre-emptive sweeps of blocks to prevent crime.
“This is not magic; there are limitations, but we tested it, and it works very well,” said Chattopadhyay.
Rather, the algorithm should be added to the toolkit of urban policy and policing strategies for tackling crime, the scientist contends:
“We have created a digital twin of the urban environment. If you feed it data about what happened in the past, it will tell you about the future,” he said.
The research team also examined police responses to crime, analysing arrest counts after incidents and comparing metrics across districts.
They found that rises in crime in wealthier districts were associated with more arrests. In disadvantaged neighbourhoods the pattern was reversed, which may indicate a bias in police response, the researchers say.
The tool was tested with historical Chicago data. The model also performed well when applied to data from seven other U.S. cities: Atlanta, Austin, Detroit, Los Angeles, Philadelphia, Portland, and San Francisco.
Earlier, in August 2021 the Pentagon tested a decision-making system, capable of foreseeing events days ahead.
In September DeepMind developed a weather-prediction algorithm.
Subscribe to ForkLog AI on TikTok!