This is the Team “MCHawks” performance analysis submitted to the U.S. Justice Department’s National Institute of Justice’s (NIJ’s) Recidivism Forecasting Challenge.
The goal of the Forecasting Challenge was to have each participating team develop a model to predict future recidivism among a sample of persons released from prison to parole in Georgia, with relevant data provided by NIJ. The primary outcome for the Challenge was prediction of re-arrest for any crime at 1, 2, and 3 years after release. A secondary outcome was the development of a “fair and accurate” model that balanced false-positive predictions between Black and White parolees. The MCHawks team’s prediction model placed in the following categories: 1) First Place: Year 2, Male Parolees; 2) First Place: Year 2, Average Accuracy; and 3) Second Place: Year 2, Female Parolees. The team also placed second, third, and fifth among the “Accounting for Racial Bias Category” in the Year 1 and Year 2 categories. This report provides details on the development of the team’s models, discusses the features used in building the predictions, and the interpretation of the results. In discussing future considerations, the report addresses how predictive models can best be applied and how subsequent similar competitions can better guide the development of fair and accurate algorithms.