This report outlines the steps taken and modeling results for the U.S. Department of Justice’s National Institute of Justice’s (NIJ’s) Recidivism Challenge.
Overall, DataRobot’sML solution was used with some basic preprocessing code written in R to arrive at the best model submission described in this report. It is noted that this type of modeling exercise would require great care if put into production. Ethical and responsible AI means fully understanding the stakeholder impact, system’s limitations, documenting risks, and most importantly risk-mitigation based on severity and likelihood to protect vulnerable populations. The model described is not intended for production based on the ethical considerations and impacts that could be amplified without careful discretion. The extent of the model’s utility is to understand and provide insights based on the data provided specifically in terms of accuracy and its relationship to protected features, race, and gender, according to proportional parity.
Downloads
Similar Publications
- Identifying Subtypes of Image-Based Sexual Abuse and their Distinct Incident Characteristics, Victimization Histories, and Personal Correlates
- The Declining Significance of Delinquent Labels in Disadvantaged Urban Communities
- A Short Form of the Measure of Adolescent Relationship Harassment and Abuse (MARSHA-SF)