Crime Prediction Using Machine Learning: A Systematic Review and Future Directions

Abstract

The increase in urban crime has promoted the use of Machine Learning as a predictive approach to support crime prevention and decision-making in public safety. This systematic review aimed to analyze the application of Machine Learning in Crime Prediction by identifying technologies, application areas, editorial quality, conceptual foundations, and predominant thematic categories. The Kitchenham and Charters methodology was followed through searches in Scopus, IEEE Xplore, Web of Science, EBSCOhost, and ScienceDirect, with the search closed on July 12, 2025. After applying exclusion criteria and quality assessment, a final corpus of 60 papers was consolidated, and its data were organized and synthesized quantitatively and critically. The results show that Python was the predominant language, used in 75% of the studies, while applications were mainly concentrated in transportation, mental health, and public safety. Likewise, publications in Q1 and Q2 journals predominated, evidencing a corpus with adequate scientific visibility. Conceptually, Machine Learning was mainly defined from a technical perspective, focused on algorithms and computational models. Finally, Cybercrime Forensics and Forensic Analytics emerged as motor themes, although several frequent lines still require greater theoretical and methodological consolidation.