XAI-Enhanced Machine Learning for Obesity Risk Classification: A Stacking Approach With LIME Explanations

Abstract

Obesity remains a critical global health challenge, necessitating early risk assessment to guide preventive measures and mitigate potential complications. While various research endeavors have explored obesity classification, many existing approaches lack reliability due to limited integration with explainable artificial intelligence (XAI) methodologies. In this study, we propose a robust machine learning framework that incorporates Explainable AI (XAI) principles to accurately estimate obesity levels and provide insights into the factors influencing the predictions. We utilize the publicly available dataset from Palechor and Manotas available in the UCI ML repository which contains relevant information on individuals' physical characteristics and behaviors. Our proposed model employs an ensemble approach, specifically a stacking algorithm, where the base estimators include the Light Gradient Boosting Machine (LGBM) classifier, the Logistic Regression (LR) classifier, and the Random Forest (RF) Classifier, and the Stochastic Gradient Descent (SGD) classifier is selected as the final estimator. To enhance model interpretability and reliability, we integrate a widely accepted XAI method, Local Interpretable Model-agnostic Explanations (LIME). Our proposed framework achieves a peak accuracy of 98.82%, surpassing most existing techniques. By incorporating LIME, we not only enhance model trustworthiness but also provide deeper insights into the factors driving obesity risk. Overall, our approach contributes to advancing personalized interventions and bridging the gap between model complexity and human understanding.