ARCHIVES

Obesity has emerged as one of the most urgent public health challenges of our time, and identifying its contributing risk factors at an early stage is vital for preventing the wide range of associated complications. Despite considerable research directed at classifying obesity levels, many existing approaches fall short in terms of consistency and trustworthiness, largely because they rarely incorporate explainable artificial intelligence (XAI). This work presents a machine learning framework that brings together predictive accuracy and model transparency through the use of XAI techniques. Our model is trained on a widely-used dataset assembled by Palechor and Manotas, available through the UCI Machine Learning Repository, which captures both physical measurements and self-reported lifestyle behaviours of individuals. At the heart of our methodology is a stacking-based ensemble that draws on the complementary strengths of four base classifiers — Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA), XGBoost, and Multilayer Perceptron (MLP) — whose combined outputs feed into a meta-learner for the final classification. To shed light on how the model arrives at its decisions, we apply two well-regarded XAI methods: Local Interpretable Model-agnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP). Together, these tools make feature contributions visible and interpretable, which is especially valuable in health-related settings where clinicians need to understand and trust model outputs. The result is a framework that not only performs well but also supports more informed, personalised obesity risk management.