Hierarchical Superiority of Machine Learning Ensembles in Diabetes Risk Apprehension: A Granular Analytical Perspective

Diabetes is a prevalent chronic disease that significantly impacts public health, contributing to increased morbidity and healthcare costs worldwide. Accurate prediction of diabetes risk is crucial for early intervention and effective management strategies. This study evaluates the performance of seven advanced machine learning models-ExtraTrees, RandomForest, XGBoost, LightGBM, Gradient Boosting, AdaBoost, and Logistic Regression-in predicting diabetes using a comprehensive dataset of health and demographic information from 100,000 individuals. The models were assessed based on accuracy, precision, recall, F1-score, and Area Under the Curve (AUC) metrics. ExtraTrees achieved the highest accuracy of 94.26%, with a recall of 94.32% and an AUC of 98.62%, demonstrating superior performance in identifying diabetes cases. However, RandomForest also performed exceptionally well with an accuracy of 93.93% and the highest AUC of 98.70%, making it a strong alternative model. XGBoost, LightGBM, and Gradient Boosting displayed balanced performance with accuracies of 90.65%, 89.62%, and 88.54%, respectively, indicating their potential in predictive modeling. AdaBoost and Logistic Regression, while demonstrating solid performance, showed slightly lower accuracy scores of 87.03% and 86.60%, respectively. These results underscore the potential of treebased ensemble models like ExtraTrees and RandomForest for precise diabetes prediction. Future research should focus on optimizing these models and exploring their deployment in clinical settings to enhance early detection, reduce complications, and improve patient outcomes.