Obesity Risk Prediction Using Fusion Ensembling Methods

Resumo

Obesity is a growing global health concern, and early risk prediction plays a vital role in enabling timely intervention. This study presents a novel hybrid ensemble model for obesity risk prediction using a dataset containing 2,111 samples and 17 features related to demographic, anthropometric, and lifestyle factors. The proposed model integrates Bagging through Random Forest and Boosting techniques using XGBoost, LightGBM, and CatBoost. These models are trained independently, and their outputs are combined using a simple averaging strategy to enhance prediction accuracy while reducing overfitting and bias. The hybrid model achieved an accuracy of 97.85%, significantly outperforming traditional models such as
Logistic Regression (86%), KNN (80%), and SVM (88%), as well as standalone ensemble methods. Comprehensive preprocessing ensured balanced classes and preserved meaningful outliers, and hyperparameter tuning was employed to optimize each base model. Additionally, feature importance analysis revealed key predictors, including meal frequency, physical activity, and water intake. This model demonstrates strong potential for use in clinical decision support systems and public health monitoring tools. The results highlight the effectiveness
of fusion ensembling in handling complex classification problems in healthcare datasets.