Interpretable EEG-Driven Diagnostic Model for Personalized Mental Healthcare Using IoT-Based WBAN Systems

Clinical diagnoses of mental health disorders are often subjective, leading to misdiagnoses and underdiagnoses. AI-based diagnostic tools have emerged as promising approaches for improving diagnostic accuracy. However, effective models with interpretable outputs remain limited, which hinders their clinical adoption. To address this gap, we developed RXGBoost, a diagnostic framework that uses clinically obtained and publicly available electroencephalography (EEG) recordings together with demographic features for personalization. It integrates Boruta-XGBoost feature selection, adaptive class weighting, and SHAP-based interpretability. In a two-stage evaluation, RXGBoost consistently outperformed deep learning models (RCNN, RLSTM, and RBiLSTM) and baseline classifiers across six disorders in accuracy, precision, sensitivity, specificity, F1-score, area under the curve, false negative rate, and false positive rate. For example, it achieved F1-score gains of 3–36% over deep models and 2–14% over baselines. Beyond performance, RXGBoost enhances clinical trust by providing transparent feature-level explanations aligned with the established neurophysiological markers. Its SHAP waterfall visualizations further clarify patient-specific diagnostic decisions, reduce uncertainty, and support timely personalized mental health care. These results highlight RXGBoost’s clinically favorable error profile and explainable decision-making, positioning it as a strong candidate for integration into IoT-based wireless body area network (WBAN) systems and clinical decision-support platforms. Future work will involve pilot studies using wearable EEG sensors in hospital and outpatient settings, followed by multi-site trials to assess its generalizability. RXGBoost can be integrated into WBAN gateways or mobile applications, enabling real-time diagnostic alerts and SHAP-based explanations for clinicians. This pathway supports transparent, personalized decision-making within remote monitoring workflows and provides a foundation for regulatory approval and adoption.