Research pick: Cruise through crime prediction with AI - "Multidimensional crime prediction technique optimisation combining feature extraction and GAN"

Researchers have developed an artificial intelligence model that predicts crime more accurately than several existing approaches by combining information about where crimes occur, when they happen, and wider social patterns. They report details of the approach in the International Journal of Innovative Computing and Applications.

The model combines a graph convolutional network, which identifies relationships between locations, with a transformer, an AI architecture designed to detect patterns over time. Together, the techniques allow the system to capture both spatial and temporal trends in criminal activity. The researchers also incorporated a generative adversarial network (GAN), a system in which two AI models compete to improve performance. The GAN was enhanced using a variational autoencoder, a method that helps generate more representative data while reducing common training problems such as biased outputs and vanishing gradients, where learning slows or stops.

The system integrates several machine-learning techniques to analyse complex datasets that traditional methods often struggle to process. In tests on historical data from several US cities, including Los Angeles and Seattle, the model achieved an accuracy rate of 86.3 per cent when predicting robberies. The strongest competing systems had an accuracy of 83.2 per cent. The new system also gave strong results across other crime categories.

The researchers suggest that accurate forecasting could help law enforcement allocate resources more effectively and identify areas at higher risk of crime. However, there are limitations. The system was less accurate in areas with sparse crime data and struggled to make predictions in locations with little or no historical information. Future work will focus on adapting the model to such environments through transfer learning, which will allow knowledge gained in one setting to be applied to another.

Xie, M. (2026) ‘Multidimensional crime prediction technique optimisation combining feature extraction and GAN’, Int. J. Innovative Computing and Applications, Vol. 15, No. 5, pp.289–299.