The metabolic illness known as diabetes is categorized by raised blood sugar points, which may lead to symptoms such as polyuria, polydipsia, and polyphagia. Diabetes is more common than other metabolic disorders. If unmanaged, it can result in severe acute and long-term complications. In order to anticipate the probability of diabetes-related complications, this research investigates the use of ensemble learning techniques such as XGBoost, AdaBoost, and Bagging. The objective is to enhance predictive accuracy using commonly available laboratory results. Performance metrics, including recall, accuracy, precision, and F1 score, were used to assess the models. The results indicate that AdaBoost outperforms the other models, achieving the highest accuracy of 94.6%, precision of 93.2%, recall 92.1%, and F 1 score of 92.6%. XGBoost also demonstrates strong performance with 94.3% accuracy, 93.1% precision, 90.1% recall, and 91.5% F1 score. Bagging, while effective, shows relatively lower performance metrics. The high predictive capability of the proposed models, particularly AdaBoost, highlights their potential for integration into online tools to assist physicians in predicting future diabetes risk and implementing preventive interventions. AdaBoost, with its balanced and superior performance, is recommended for applications where high classification accuracy is essential, while XGBoost remains a strong alternative. Bagging may be preferred in scenarios prioritizing model simplicity and interpretability. © 2024 Elsevier B.V., All rights reserved.