Machine Learning–Based Obesity Detection Using Feature-Optimized Xgboost And Comprehensive Evaluation

Abstract

Obesity has emerged as a major global health concern due to its strong association with chronic diseases such as diabetes, hypertension, cardiovascular disorders, and metabolic complications. Early and accurate prediction of obesity risk is essential for guiding preventive interventions. This project presents a robust machine learning framework for classifying obesity levels using an enhanced XGBoost model trained on a structured lifestyle and physical-condition dataset. The dataset undergoes systematic pre-processing that includes label encoding, normalization, and stratified train-test splitting to ensure reliable learning. The XGBoost classifier is chosen for its superior ability to capture complex feature interactions, handle mixed data types, and reduce over fitting. Comprehensive evaluation metrics such as accuracy, confusion matrix, and classification report demonstrate that the model achieves high predictive performance. Important visualizations, including correlation heatmaps and feature-importance plots, provide deeper insights into the factors influencing obesity outcomes.

The final trained model and processed datasets are saved for deployment and future research. This study highlights the effectiveness of gradient-boosting approaches in health-risk prediction and contributes to the development of intelligent decision-support systems in healthcare.