Machine Learning and Risk Assessment: Random Forest Does Not Outperform Logistic Regression in the Prediction of...

Area of Science:

• Forensic Psychology
• Criminology
• Machine Learning in Behavioral Science

Background:

• Actuarial risk assessment instruments (ARAIs) are crucial for predicting recidivism in sexual offenders, yet improvements in predictive performance are sought.
• Machine learning (ML) algorithms offer potential for detecting complex patterns and nonlinear effects in risk factor data.

Purpose of the Study:

• To compare the predictive performance of conventional logistic regression with the random forest ML algorithm for sexual offender recidivism.
• To investigate potential nonlinear effects in risk factors using interpretable ML methods.

Main Methods:

• A prospective-longitudinal study followed 511 adult male sexual offenders for an average of 8.2 years.
• Logistic regression and random forest algorithms were used to analyze data, incorporating Static-99 and Stable-2007 risk factors.

Main Results:

• The random forest algorithm did not demonstrate superior predictive performance compared to logistic regression.
• Interpretable ML methods did not reveal robust nonlinear effects between risk factors and recidivism.

Conclusions:

• Logistic regression remains a statistically sound and effective method for developing and clinically applying ARAIs for sexual offenders.
• The study supports the continued use of logistic regression in actuarial risk assessment due to comparable performance and established interpretability.