Know the Unknown or Better Know the Known: A Machine-Learning Approach to Risk Assessment for Sexual Offending

• In Event: Community and Institutional Responses to Sexual Offending

Abstract

For the past several decades, much research has gone into the discovery of risk factors and protective factors in explaining sexual offending. However, little effort has been made to review how much we actually know about the prediction of sexual offending and how to maximize the utility of such knowledge. By harnessing machine learning algorithms, this study aims to advance current practice in developing risk assessment tools. Machine learning is an iterative process in which a system optimizes prediction based on previous examples. It holds promise for increasing the power and efficiency of criminal justice risk assessment and particularly for the risk of sexual re-offending in which “the low base rate” has always been a problem. The study will also offer in-depth analysis on the overestimation of risk and its implications for policy and research.

Authors

• Erika Parks, Urban Institute

• KiDeuk Kim, Urban Institute