Performance Benchmarking of Optimized Stacked Ensemble Frameworks in Blood Glucose Level ( BGL ) Prediction

This paper presents a study of different stacked ensemble techniques for long‐term Blood Glucose Level (BGL) prediction in adults. Stacked ensemble models were developed by combining various machine learning techniques, namely eXtreme Gradient Boosting (XGBoost), Adaptive Boosting (AdaBoost), and Random Forest (RF). These machine learning models served as base learners for a Linear Regression (LR) based meta‐learner model. The models were evaluated on 30% held‐out test samples using a suite of different evaluation metrics, incorporating 10‐Fold Time Series Cross‐Validation (TSCV). The XGBoost technique was effectively used for feature selection. The hyperparameters were optimized using the Grid Search (GS) and Bayesian Optimization (BO) techniques. The findings highlight a superior performance of the ensemble model optimized using a BO approach. The results confirmed the superior performance of the ensemble model designed using XGBoost and AdaBoost techniques as base learners for the LR model, incorporating the BO approach for hyperparameters optimization.