This report is a response to the National Institute of Justice’s (NIJ’s) Recidivism Forecasting Challenge 2021, which aimed to increase public safety and improve the fair administration of justice across the United States.
The models and procedures outlined in this report were judged to be the third best performance in the Challenge for predicting recidivism in Year 1 for male parolees, female parolees, and on average accuracy. This report outlines the process of developing such a forecasting model. Two main algorithms were used for developing a prediction model in this effort. The first type of model is a simple penalized regression model. The second type of model is Extreme Gradient Boosting, XGBoost, a highly effective gradient tree-boosting algorithm that has been demonstrated in many data science competitions. Both modeling approaches typically provide optimal solutions for rectangular data. The datasets provided by NIJ for the Challenge are described, along with supplemental datasets compiled by the researcher.
Downloads
Similar Publications
- Inequalities in Exposure to Firearm Violence by Race, Sex, and Birth Cohort From Childhood to Age 40 Years, 1995-2021
- Exposure to Violence at School and School-Related Promotive Factors Among Adolescents in Urban Settings
- Child sexual abuse images and youth produced images: The varieties of Image-based Sexual Exploitation and Abuse of Children