Enhancing Korean adolescent suicide risk prediction with TabR: A generative AI and explainable retrieval-based deep...

Area of Science:

• Public Health
• Artificial Intelligence
• Adolescent Psychology

Background:

• Adolescent suicide is a critical public health issue in South Korea, with high youth suicide rates.
• Leading causes include academic stress, mental health issues, and societal pressures.
• Early identification of at-risk adolescents is vital for effective intervention.

Purpose of the Study:

• To develop and evaluate a novel deep learning model, TabR, for predicting adolescent suicide attempts.
• To address challenges of class imbalance and limited positive cases in suicide prediction datasets.
• To compare TabR's performance against established machine learning and deep learning models.

Main Methods:

• Utilized the Korea Youth Risk Behavior Survey dataset (2020-2022) with 161,646 participants.
• Employed a tabular variational autoencoder (TVAE) for synthetic data augmentation to handle class imbalance.
• Developed and tested TabR, a retrieval-augmented residual neural architecture, comparing it with 8 baseline models.

Main Results:

• TabR, trained on TVAE-augmented data, achieved superior performance.
• Achieved an accuracy of 0.9726, an Area Under the Curve (AUC) of 0.8963, and an F1-score of 0.4754.
• Outperformed all 8 baseline models in suicide attempt prediction.

Conclusions:

• TabR demonstrates significant potential for improving the prediction of adolescent suicide attempts.
• The study highlights the effectiveness of data augmentation techniques for imbalanced datasets in mental health research.
• Further research is needed for external validation and clinical impact assessment before deployment.