This report presents the features and results of the National Institute of Justice Recidivism Forecasting Challenge (the Challenge).
The Challenge was a research competition that required entrants to develop and train software models to forecast recidivism for individuals released on parole from the state of Georgia. Entrants were provided a dataset that enabled them to train their forecasting models by exploring gender, racial, and age differences for individuals on parole, as well as other information. The Georgia Department of Community Supervision was selected as a partner in administering the Challenge on the strength of prior state-funded investments that improved the breadth of state data-collection and data- sharing capabilities. The Challenge attracted just over 70 teams with a variety of expertise and access to resources. The major sections of this report address 1) the Challenge design and judging criteria; 2) the models and methods used for contextualizing and comparing competitors’ data selection, analyses, and conclusions; 3) the results of the Challenge; and 4) the conclusion of the Challenge and next steps. The report concludes that the winning forecasts performed substantially better than random chance and naïve demographic models. The differences in accuracy between the winning and naïve models were attributed to the use of more advanced statistical techniques, e.g., regression, random forest, neural networks; and the incorporation of additional data from the Georgia Department of Community Supervision beyond the demographics used in the naive models. Remaining research issues to be addressed by NIJ in assessing risk of recidivism are noted.