This report explains how the author addressed issues posed in the National Institute of Justice’s (NIJ’s) most recent (2021) Recidivism Forecasting Challenge.
The author - Jianye Ge of the University of North Texas Health Science Center – submitted the forecast likelihoods as a small team and was one of the winners in Year 1. The current report explains how the data were analyzed in the Challenge. The section on methods addresses the data and variables provided by NIJ on the parolees who were assessed for risk of recidivism. The data provided by NIJ included both training and test data. Multiple ways were tried to improve the overall performance, such as changing the classifier threshold from 0.5 to other numbers, replicating more true cases to balance they number of false and true cases. None of them had any substantial improvement to the Brier Scores. The author notes she should have separated the males and females in predictions, since the recidivism patterns for males and females could be different and the challenge did determine the winners for the male and female parolees.
Downloads
Similar Publications
- Machine Learning and the Prevention of Mass Shooting in the United States
- Enhancing Fault Ride-Through Capacity of DFIG-Based WPs by Adaptive Backstepping Command Using Parametric Estimation in Non-Linear Forward Power Controller Design
- Solving Cases of Sudden Unexpected Natural Death in the Young through Comprehensive Postmortem Genetic Testing