AI-driven crime prediction: a systematic literature review

The topic of crime analysis and prediction is crucial for public safety, and advanced AI has shown promise in enhancing this research area. Exploring these advanced AI-based schemes and summarizing them in one place can help communities such as policymakers, law enforcement agencies, and researchers. In this direction, literature presents some comprehensive state-of-the-art studies, however, these studies lack a thorough review of the latest AI methods used in crime prediction. This systematic literature review aims to fill these gaps by exploring recent advancements in Machine Learning, especially in Deep Learning, for crime prediction. This study delves into various digital repositories to extract relevant research articles using a robust qualitative selection process. A total of 55 top-quality research articles have been analyzed in terms of publishing venue, employed predictive model, geographical area for experiments, use of real-time data, data source, and time spans for crime records. The detailed analysis reveals that IEEE Access is a top publishing venue for crime prediction-related articles. In the dimension of applied techniques, traditional methods persist (33%), while innovative hybrid models and emerging techniques such as Convolutional Neural Networks (CNNs), Long Short-Term Memory Networks (LSTMs), Generative Adversarial Networks (GANs), Graph Neural Networks (GNNs) show a trend toward diverse integration. The geographical focus primarily centers on U.S. cities, appearing in 56% of the papers, highlighting global disparities. Additionally, there is a significant gap in incorporating real-time data in crime prediction models, underscoring the need for future exploration in this direction. These findings provide valuable insights for researchers, policymakers, and law enforcement agencies.