Classification of Adolescent Drinking via Behavioral, Biological, and Environmental Features: A Machine Learning...

Area of Science:

• Neuroscience
• Public Health
• Machine Learning

Background:

• Adolescent alcohol use is a significant public health concern, linked to neurodevelopmental issues and future mental health disorders.
• Existing classification models often rely on costly neuroimaging, limiting large-scale application.
• Clinical measures offer a practical, accessible alternative for screening adolescent alcohol consumption.

Purpose of the Study:

• To develop and validate a clinical-only machine learning framework for classifying adolescent alcohol use.
• To address limitations of prior studies, including focus on adults, inadequate handling of confounders, and class imbalance.
• To enhance early identification and intervention for adolescent alcohol use through accessible screening methods.

Main Methods:

• Proposed FocalTab, integrating TabPFN with focal loss to mitigate class imbalance and improve generalization.
• Implemented a preprocessing step to remove confounding factors like age and substance use.
• Compared FocalTab against state-of-the-art methods across various dataset settings, including stringent conditions excluding confounders.

Main Results:

• FocalTab achieved the highest accuracy (84.3%) and specificity (80.0%) in the most challenging setting (excluding age and substance use).
• Competing models showed significantly lower specificity (12-24%) under the same stringent conditions.
• SHapley Additive exPlanations (SHAP) identified key clinical predictors, supporting clinical utility.

Conclusions:

• FocalTab demonstrates robust performance in classifying adolescent alcohol use using only clinical data.
• The framework effectively addresses class imbalance and confounding factors, outperforming existing methods.
• Findings support the use of FocalTab for scalable, cost-effective screening and early intervention in adolescents.