This report on the winner of the U.S. Justice Department’s National Institute of Justice’s (NIJ’s) Recidivism Forecasting Challenge with Multi-Target Ensembles pertains to male, female, and overall categories in the first year.
Classification- based quantitative data are primarily about feature engineering and model ensembling. The former enriches the existing patterns in the data, and the latter produced robust predictions even from a limited amount of data. The winning solution for the Recidivism Forecasting Challenge included heavy amounts of each. A roadmap to this solution is provided, along with source code and guidance on how to produce similar results on future datasets. Machine learning models are clearly up to the task of predicting recidivism for the purposes of parole judgments or monitoring. The predicted probabilities spanned the full range from near-zero to very likely.