Sign In

Communications of the ACM

ACM TechNews

Opening the Lid on Criminal Sentencing Software


Handcuffs, ready for repeat offenders.

A team of researchers at Duke University has developed the Supersparse Linear Integer Model, a machine learning alrorithm that produced a model to predict whether a convicted criminal will commit another crime if released.

Credit: today.duke.edu

Researchers at Duke University are using machine learning to train computers to build statistical models that can predict future criminal behavior, which are just as accurate as other methods but more transparent and easier to interpret.

The team developed the Supersparse Linear Integer Model (SLIM) using a public dataset of more than 33,000 inmates released from prison in 15 states in 1994 and tracked for three years.

SLIM scanned the data to seek patterns, accounting for factors such as gender, age, criminal history, and dozens of other variables in order to predict future offenses. The system then produced a model to predict whether a defendant will relapse or not, based on those same rules.

The researchers also developed another machine-learning algorithm, CORELS, which digests data about new offenders, compares them to past offenders with similar characteristics, and divides them into groups to help predict how they might behave in the future.

From Duke Today
View Full Article

 

Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA


 

No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account