Predicting recidivism among youth offenders: Augmenting professional judgement with machine learning algorithms

Area of Science:

• Criminology
• Data Science
• Machine Learning

Background:

• Offender rehabilitation aims to reduce recidivism.
• Singaporean probation officers use experience and risk assessment tools for sentencing recommendations.
• Maximizing data utility for accurate recidivism prediction is challenging with current methods.

Purpose of the Study:

• To explore the effectiveness of predictive modeling using statistical learning methods for recidivism prediction.
• To improve the accuracy of recidivism prediction by analyzing discrete-level administrative data.

Main Methods:

• A random forests model was developed using data from 3744 youth offenders.
• The model utilized administrative data, including socio-economic factors, without prior assumptions on predictor influence.
• Sixty percent of data was used for model development and 40% for testing.

Main Results:

• The random forests model achieved a classification accuracy of approximately 65% and an Area Under the Curve (AUC) of 0.69.
• This predictive model outperformed existing models that analyzed aggregated data using conventional statistical methods.

Conclusions:

• Statistical learning methods applied to discrete-level administrative data offer higher accuracy in predicting recidivism.
• This enhanced predictive capability allows for targeted intervention efforts towards individuals at higher risk of reoffending.