Researchers have developed a new artificial intelligence (AI) system that can predict robberies and other crimes more accurately than several existing methods.
By studying where crimes happen, when they occur and broader social patterns, the system achieved an accuracy rate of 86.3% when forecasting robberies in several U.S. cities.
Predicting crime is extremely challenging because criminal activity is influenced by many different factors.
Crimes do not happen randomly. Certain locations may have higher crime rates, and criminal activity can also follow patterns related to time, such as specific days, seasons or social events.
To tackle this problem, the researchers combined several advanced AI techniques into a single system.
One part of the model examines how different locations are connected and identifies areas that may share similar crime patterns. Another part studies changes over time, helping the system detect trends and predict when crimes are more likely to occur.
The researchers also added a technique known as a generative adversarial network, or GAN. In a GAN, two AI models essentially compete with each other, allowing the system to improve its ability to recognize patterns and make predictions.
They further enhanced the system with another machine-learning method called a variational autoencoder, which helps create more realistic training data and reduces common problems that can make AI systems less reliable.
Together, these technologies allow the new system to process large and complicated data sets that are often difficult for traditional methods to analyze.
The researchers tested their model using historical crime data from several American cities, including Los Angeles and Seattle. When predicting robberies, the new system correctly forecast crime patterns 86.3% of the time. By comparison, the strongest competing models achieved an accuracy rate of 83.2%.
The system also performed well when analyzing other types of crime, suggesting that the approach may have broad applications in crime forecasting.
The researchers believe such tools could help law enforcement agencies use their resources more efficiently. If police departments can identify areas that may face a higher risk of crime, they may be able to plan patrols and prevention strategies more effectively.
However, the technology still has limitations. The model was less accurate in places where little crime data existed. It also struggled to make predictions in areas with very limited or no historical records.
To address this challenge, future research will focus on transfer learning, a technique that allows an AI system to apply knowledge learned in one location to another. This could help improve predictions in places where historical crime information is scarce.
While AI cannot predict every crime, the study shows that advanced machine-learning systems may become valuable tools for understanding crime patterns and supporting more informed decision-making in public safety.
