The algorithm with an accuracy of 300 meters allows you to predict where an attack or theft will be committed a week before it happens. Although such predictive models can strengthen the power of the state through illegal surveillance of innocent people, at the same time they allow surveillance of the state, revealing systemic bias in the actions of law enforcement agencies.
Achievements in the field of machine learning and artificial intelligence have aroused considerable interest among governments of different countries. And their interest is understandable: if there was a working tool for predicting crimes, it would greatly simplify the work of law enforcement agencies and in the future dramatically reduce the level of street crime. Naked Science talked about one of these models, which gives weekly predictions about terrorist attacks based on data only from open sources, last year.
However, most of the previous attempts to predict crime were quite contradictory and inaccurate. Mainly because they often used the so-called epidemic or seismic approach, when crime occurs in certain “hot spots”, which then spread to nearby areas. At the same time, the complex social environment of cities and their natural topology are overlooked, the relationship between crime and the consequences of police coercion is not taken into account.
Data analysts and sociologists from the University of Chicago (USA) have developed a new algorithm that predicts crime by studying patterns in time and geographical distribution of violent crimes (murder, assault, battery, etc.) and crimes against property (burglary, ordinary street theft and car theft, etc.), using only publicly available data. The model can make forecasts of future crimes for a week ahead with an accuracy of about 90%. The scientists described their stochastic inference algorithm in an article published in the journal Nature Human Behavior.
The new model divides the city into identical squares with a side of about 300 meters, analyzes the time and place of individual crimes and identifies patterns for predicting future events. Initially, the model was tested on data on assaults and thefts in the third most populous city in the United States of America — Chicago. However, the model worked just as well with data from seven other American cities: Atlanta, Austin, Detroit, Los Angeles, Philadelphia, Portland and San Francisco.
As part of a separate predictive model, the research team studied the response and actions of the police to crimes in various parts of the city, analyzing the number of arrests after the relevant incidents and comparing these indicators among districts with different socio-economic status. The authors of the work noticed that an increase in the crime rate in richer areas leads to more arrests in them, while the number of arrests in disadvantaged areas is decreasing. However, a similar increase in the number of crimes in poor areas does not lead to the expected increase in the number of arrests there, which indicates bias in the police response and law enforcement.
And yet, despite the high accuracy of their crime prediction model, scientists note that it should not be used directly to ensure law and order. After all, an increase in the number of police officers in those areas of the city where a crime is expected will lead to a change in the modeling conditions and will only reduce the effectiveness and accuracy of prediction. Instead, the model should be added to the set of urban policy tools and police strategies to combat crime.
We have created a digital twin of the urban environment.
If you provide him with data about what happened in the past, he will tell you what will happen in the future. It’s not magic, there are limitations, but we tested the model and it works very well. Now you can use it as a modeling tool to see what happens if crime increases in one area of the city or law enforcement increases in another area. If you use all these variables, you can see how systems evolve in response,” summed up Ishanu Chattopadhyay, associate professor at the University of Chicago Faculty of Medicine and senior author of the new study.