SH-BRB: A semantically-grounded and explainable AI framework for dynamic mental health assessment

Abstract High-performance artificial intelligence models in mental health assessment often lack transparency, while traditional expert systems rely heavily on subjective knowledge. Additionally, statistical theoretical models show limited effectiveness in individual-level prediction. This study aims to propose a high-performance and inherently interpretable dynamic assessment framework, SH-BRB, to resolve the trade-off between predictive accuracy and interpretability in clinical screening.First, an automated knowledge-transformation protocol is introduced to convert the validated factor structure and path coefficients of Structural Equation Modeling (SEM) into the hierarchical topology and initial parameters of a Belief Rule Base (BRB), ensuring objective initialization. Second, a context-aware dynamic reasoning mechanism is designed to simulate clinical situational awareness by adaptively modulating rule weights and decision thresholds based on individual medical history. Comparative experiments demonstrate that SH-BRB outperforms state-of-the-art models, such as XGBoost and CatBoost, under identical conditions. Specifically, the framework achieves a 99.3% precision in identifying high-risk individuals. Furthermore, randomized sampling experiments validate the model’s transparency and decision traceability, ensuring clinical logic consistency. The proposed SH-BRB framework effectively integrates statistical rigor with transparent reasoning. It overcomes the limitations of "black-box" models and subjective expert systems, providing a trustworthy and verifiable decision-support tool for dynamic mental health assessment.