Developing SHAP interpretable machine learning models for assessing biopsychosocial risk in female drug users: a...

Area of Science:

• Public Health
• Machine Learning
• Psychology

Background:

• Comprehensive risk assessment tools for female drug users are lacking.
• Existing tools often fail to integrate biopsychosocial dimensions.
• There is a critical public health need for objective risk assessment in this population.

Purpose of the Study:

• To develop a comprehensive machine learning model for assessing biopsychosocial risks in female drug users.
• To evaluate physiological, psychological, social, and self-control dimensions.
• To identify key variables contributing to overall risk.

Main Methods:

• A biopsychosocial model was applied, collecting data on five dimensions from 96 participants.
• Machine learning classifiers were trained and evaluated using oversampled data.
• SHAP (SHapley Additive exPlanations) was used for model interpretability.

Main Results:

• Machine learning models showed good performance in classifying risks across multiple dimensions.
• Oversampling improved classification accuracy for most dimensions and total risk.
• Random Forest and Logistic Regression were identified as optimal classifiers; SHAP revealed key risk variables.

Conclusions:

• The developed machine learning model effectively identifies at-risk female drug users.
• The model integrates physiological, psychological, cognitive, drug refusal, social support, and self-control dimensions.
• Future research will expand the sample size to enhance model generalizability and reduce overfitting.