NCJ Number
252682
Date Published
July 2018
Length
4 pages
Annotation
This policy digest summarizes and discusses policy implications of a 2017 study of machine learning applied to pretrial release decisions.
Abstract
In order to make pretrial release decisions, judges must decide whether to detain people who are facing criminal changes in jail or allow them to go home until their trial based on the available information. In this decision, the judges use the information, such as prior criminal history and the nature of the alleged offense, to assess if a person is likely to appear for his or her future trial, known as "flight risk," or commit a crime before the trial. This policy digest presents that it may be possible to improve judges' pretrial decision-making with the help of machine learning. The author's present findings that machine learning models outperform the current practice of judges by more accurately predicting who will fail to appear for trial. It was found that using the machine learning predictions of flight risk instead of judge decisions to select people to detain pretrial could reduce failures to appear by 24.7% and all other crime types by 11.1% with no change in pre-trial detention rates, or reduce pre-trial detention populations by 42.0% with no change in the failure to appear rate.