Improving Obesity Classification Accuracy Using SMOTE-based Dataset Balancing and Ensemble Machine Learning Approaches

Obesity categorization is a public health issue, especially when unbalanced datasets impair model performance. For obesity categorization, this research found XGBoost, CatBoost, and AdaBoost to be the most accurate machine-learning(ML) models. Stacking ensembles of these three models yielded 87.3% prediction accuracy. This research also uses the Synthetic Minority Over-sampling Technique (SMOTE) to balance the dataset by over-sampling the minority class. XGBoost, Random Forest, and CatBoost performed best on the balanced dataset, with a stacking ensemble method improving accuracy to 92.5%. Data balancing strategies improve model performance and ensemble methods improve obesity classification accuracy. For future health-related predictive analytics applications, the findings emphasize the need to overcome class imbalance in machine learning tasks.