The Use of AI in Corrections and the Development of the IDRACS Risk Classification Tool

• In Event: Assessing Re-arrest Using Predictive Modeling: The Integrated Dynamic Risk Assessment for Community Supervision Tool

Abstract

Artificial intelligence (AI) applications in corrections hold the potential to transform institutional and community corrections by improving efficiency, enhancing decision-making, and reducing costs. However, the deployment must be cautious to avoid unintended consequences, especially concerning ethical considerations and data quality. The Integrated Dynamic Risk Assessment for Community Supervision (IDRACS) risk classification tool tries to enhance the accuracy of risk predictions for individuals on probation or parole in Georgia. The project developed a series of logistic regression models tailored to different supervision types and periods, incorporating both static and dynamic factors. Findings reveal that models including dynamic measures significantly improve predictive accuracy and that time-specific models provide better predictions during early supervision. Machine learning methods yield modest gains but were not substantively superior to traditional models. Post-COVID validation indicate these models remain robust in predicting recidivism, demonstrating their potential for broader application in criminal justice settings.

Authors

• Christopher Inkpen, RTI International

• Pam Lattimore, RTI International

• Nicholas Powell, Georgia Department of Community Supervision