Ensemble Learning for Healthcare: A Comparative Analysis of Hybrid Voting and Ensemble Stacking in Obesity Risk Prediction

Researchers compared ensemble machine learning techniques for predicting obesity risk, finding that ensemble stacking with a neural network meta-classifier outperformed hybrid voting methods, particularly on complex datasets. The study evaluated nine ML algorithms across 50 hyperparameter configurations, demonstrating that stacking achieves superior accuracy (up to 98.98%) for healthcare predictive modeling.

This research addresses a significant gap in healthcare machine learning by systematically comparing two popular ensemble approaches for obesity risk prediction. Obesity represents a critical public health challenge linked to multiple chronic diseases, making accurate early detection valuable for preventive medicine. The study's rigorous evaluation of 50 hyperparameter configurations across nine base algorithms provides substantial empirical evidence for practitioners selecting ensemble architectures.

The healthcare sector increasingly adopts machine learning to improve diagnostic accuracy and patient outcomes. Ensemble methods combine multiple weak learners to reduce overfitting and improve generalization—a critical consideration when deploying models to real clinical settings where data distributions vary. This research contextualizes how different ensemble paradigms handle varying data complexities, with weighted voting providing stability on simpler datasets while stacking excels on complex distributions.

For healthcare technology developers and medical institutions, these findings inform architectural decisions when building predictive systems. Organizations must weigh stacking's superior accuracy against its computational overhead and implementation complexity. The consistent performance of stacking across datasets suggests it represents a more robust choice for production deployments, despite higher training costs. Weighted voting offers practitioners a computationally efficient alternative when datasets exhibit lower complexity.

Future research should explore stacking performance on larger, real-world clinical datasets and investigate interpretability challenges—critical for healthcare applications requiring clinical validation. The framework presented enables healthcare organizations to benchmark their own obesity prediction systems and select appropriate ensemble strategies based on their specific data characteristics and computational constraints.

• →Ensemble stacking with MLP meta-classifier achieves 98.98% accuracy on complex obesity datasets, outperforming voting methods
• →Weighted voting performs optimally on simpler datasets while stacking demonstrates superior generalization on complex distributions
• →Systematic hyperparameter optimization across 50 configurations identifies optimal base learner combinations for ensemble approaches
• →Healthcare practitioners should prioritize stacking for production systems despite higher computational costs due to superior accuracy
• →Ensemble method selection should align with data complexity and institutional computational resources