Ensemble Regression and Explainable AI for Predicting Resource Utilization Efficiency in 6G-Enabled Smart Healthcare Systems

Optimal management of resources will be founda- tional to future health systems designing around 6G technology. The combination of ultra-reliable, very low-latency connectivity with autonomous AI-based decision-making will automate many of the operational aspects of healthcare delivery. The research presented here establishes a total machine learning framework that can implement various regression analysis techniques and ensemble models to predict Resource Utilization Efficiency (RUE). We constructed a diverse dataset from clinical, operational, and telecommunications-based variables and utilized multiple data preprocessing techniques (imputation, encoding, scaling, and outlier correcting) to optimize the training of our six benchmarked regression analyses: Linear Regression, Random Forest, Gradient Boost, XGBoost, Support Vector Regression, and K-Nearest Neighbors. Results demonstrated that tree-based models achieved the highest predictive accuracy, with Random Forest, Gradient Boosting, and XGBoost consistently outperforming linear and kernel-based approaches. To further enhance performance, ensemble learning methods (averaging, blending, and stacking) were employed, with stacking ensembles delivering the best overall results (MSE = 1.86 × 10−5, R2 = 0.9998). To produce robust models through hyperparameter tuning with GridSearchCV and Bayesian optimization; the SHAP analysis method was conducted to provide interpretation to the decision process, revealing that Network Performance (speed), Length Of Stay and Health Status were the most significant variables in predicting RUE. Promoting predictiveness while maintaining transparency provides a concrete, interpretable decision support tool for healthcare decision-makers. With the proposed framework, intelligent; sustainable and explainable; 6G supported Healthcare Management may continue to expand to include federated learning; real-time implementation; and multi-modal data.