A domain-informed ensemble learning framework with explainable visual analytics for obesity risk assessment

Abstract

Obesity is a multifactorial chronic disease driven by complex interactions among genetic predisposition, metabolic processes, and lifestyle behaviors, posing significant challenges for accurate risk prediction and clinical decision-making. In this study, we propose a Domain-Aware Ensemble Learning Framework for obesity risk prediction that integrates domain-specific feature engineering with explainable visual analytics. The framework transforms heterogeneous clinical and lifestyle data into clinically meaningful indicators, including the Energy Balance Index (EBI), metabolic efficiency ratios, and family lifestyle amplification scores. To enhance predictive robustness and generalization, a meta -learning strategy is implemented using a multi-level stacking ensemble architecture that combines diverse machine learning models. Experimental results demonstrate stable and reliable performance, achieving a mean accuracy of 83.7% ± 3.6% with low variance across cross-validation folds. In addition, an explainable analytics layer is introduced to improve model transparency, providing insights into feature importance and decision behavior. Feature efficiency analysis reveals that a compact subset of 14 domain-informed features accounts for over 80% of the predictive power. The proposed framework offers a robust, interpretable, and scalable solution for obesity risk assessment, supporting AI-driven personalized healthcare and clinical decision support.