Interpretable Machine Learning to Identify Risk Factors for Recidivism in Intimate Partner Violence.

Area of Science:

• Public Health
• Criminology
• Machine Learning

Background:

• Intimate Partner Violence (IPV) is a major global health concern with increasing rates.
• Predicting recidivism in IPV offenders is vital for effective prevention and intervention strategies.
• Existing risk assessment tools may lack standardization and struggle with non-linear relationships.

Purpose of the Study:

• To develop interpretable machine learning models for predicting physical assault recidivism among IPV offenders.
• To identify key features associated with recidivism risk.
• To improve the objectivity and accuracy of IPV risk assessment.

Main Methods:

• Utilized a four-year clinical study dataset.
• Applied filtered target encoding to standardize severity scores and handle non-linear associations.
• Developed interpretable machine learning models to analyze feature importance.

Main Results:

• Combining self-reported and partner-reported variables significantly enhanced predictive accuracy.
• Decreased substance use and avoiding partner contact were associated with lower recidivism risk.
• Separation processes were identified as a factor correlating with higher reoffending likelihood.

Conclusions:

• The developed machine learning models offer a more nuanced understanding of IPV recidivism risk factors.
• Findings can inform the development of more effective IPV treatment and management strategies.
• Improved risk assessment can help address disparities in IPV care and reduce reoffending rates.