Taking the Long View of Risk Assessment: The Promise and Pitfalls of Machine Learning in Actuarial Decision-Making for Criminal Justice Agencies

• In Event: Risk Assessment:  Machine Learning vs. Regression—Which Direction Should We Go

Abstract

Over the past decades, the study of machine learning has grown into a broad discipline that has produced practical applications of predictive modeling, such as identifying spam emails and recommending movies for movie streaming service subscribers. It also holds promise for actuarial decision-making in criminal justice. This study presents findings from an NIJ-funded national study on juvenile sex offenders and discusses the potential of the machine learning approach to actuarial risk assessment. We provide a review of current practice in risk assessment, as well as practical insights into how to improve it with an increasing volume of administrative data from criminal justice agencies and new advances in statistical methods.

Authors

• KiDeuk Kim, Urban Institute

• Grant Duwe, Minnesota Department of Corrections