Artificial Intelligence and Machine Learning Methods in Recidivism Prediction: A Thematic Literature Review

Recidivism remains a persistent challenge in criminal justice systems because data are fragmented across jurisdictions, privacy laws are stringent, and legal definitions of recidivism vary. This thematic literature review synthesizes evidence on risk factors influencing recidivism, traditional and centralized artificial intelligence and machine learning (AI/ML) approaches, explainable AI (XAI) for transparency and bias auditing, generative AI for local data augmentation, and federated learning in privacy-sensitive environments. The review identifies recurring patterns, including improved ensemble performance on balanced datasets and increased use of XAI for bias auditing, while also highlighting persistent limitations related to methodological variation, privacy constraints, jurisdictional heterogeneity, and fairness concerns. Based on these findings, Recid-Fed is proposed as a conceptual research agenda to guide future research on privacy-preserving and jurisdiction-aware recidivism prediction. Rather than presenting an implemented system, this paper consolidates the current literature and outlines the key research requirements that should be addressed in future framework development and empirical validation studies.