A contribution-interpretable belief rule base model for obesity risk prediction using physical indicators and dietary...

Obesity is a pressing global endocrine disorder that disrupts hormonal axes and elevates the risk of type 2 diabetes, cardiovascular disease, and metabolic conditions. Early and reliable risk prediction is crucial for timely intervention, yet most existing models fail to balance predictive accuracy with interpretability, limiting their clinical applicability.

We propose a dual-layer belief rule base model with hybrid evolutionary optimization and contribution analysis (DBRB-C) for multilevel obesity risk prediction. The framework integrates data-driven learning with knowledge-based reasoning through a hierarchical architecture, employs a differential evolution-particle swarm optimization strategy for parameter learning, and introduces a contribution belief matrix with reverse contribution analysis to trace decision-making pathways.

Evaluated on real-world datasets from Latin America and Turkey, DBRB-C achieved higher accuracy and stability than conventional machine-learning models. The model explicitly identified modifiable dietary behaviors-such as vegetable consumption frequency and number of daily meals-as key determinants of obesity risk, while providing fully traceable reasoning for each prediction.

The DBRB-C model successfully bridges the gap between accuracy and interpretability in obesity risk assessment. Its transparent, contribution-aware architecture offers a reliable foundation for personalized lifestyle interventions and supports evidence-based clinical decision-making in endocrinology and preventive health.