Explainable AI for the Prediction and Estimation of Obesity Levels using Machine Learning Models

In recent years, the global obesity rate has surged, highlighting the urgent need to address this critical public health issue. This study proposes a predictive model for obesity levels based on physical condition and eating habits, utilizing advancements in machine learning and available datasets. The Obesity level dataset from the UCI repository was balanced using Support Vector Machines Synthetic Minority Oversampling Technique (SVM-SMOTE) and standardized. Twelve classifiers are employed with Shapley Additive exPlanations (SHAP) used to interpret the best-performing models and elucidate their decision-making processes. Often seen as black boxes, artificial intelligence models were demystified in this study by using SHAP values to achieve interpretability and causality. The AdaBoost model excelled, achieving an accuracy, F1-score and precision of 89.22%. Incorporating explainability enhances transparency and helps identify the root causes of obesity, providing valuable insights into this pressing health issue. This comprehensive analysis is crucial for the ethical adoption of artificial in healthcare, addressing transparency, causality, and interpretability in diagnosis. The proposed technique demonstrates excellent performance and offers a robust framework research and applications in public health.