Opening the Black Box: Explainable AI Insights into Obesity Prediction Using SMOTE-Balanced Stacking Ensembles

Obesity, caused by an energy imbalance, is a major public health issue. Machine learning algorithms can be hampered by class imbalance in obesity dataset analysis. This research makes an attempt tobalance an unbalanced obesity dataset using the Synthetic Minority Over-sampling Technique (SMOTE) and uses this upgraded dataset for training Decision Trees, Support Vector Machines, Logistic Regression, and sophisticated boosting techniques like XGBoost and AdaBoost. Through thorough experimentation, the best three machine learning approaches are determined and a stacking ensemble model is created using these models. The proposed model achieves 92.5% accuracy on the test dataset. Explainable machine learning methods like SHAP is also used to improve forecasts' interpretability. These approaches reveal the feature contributions affecting the model's predictions, helping us understand obesity's causes. Findings show that SMOTE oversampling can alleviate class imbalance and that stacking ensemble models can improve predictive performance. Explainable AI also makes obesity research machine learning applications more transparent, enabling better public health interventions and policy decisions.