A HYBRID ENSEMBLE MACHINE LEARNING FRAMEWORK FOR EARLY PREDICTION OF TYPE 2 DIABETES: A STRUCTURED APPROACH TO PREDICTIVE HEALTHCARE

Abstract

Background: Type 2 Diabetes Mellitus (T2DM) is a significant worldwide health concern, with growing incidence and a considerable proportion of undiagnosed cases, specifically among low- and middle-income countries. Early detection of persons at risk is critical for prompt management and reducing long-term problems. Conventional diagnostic procedures continue to fall short in capturing complex, multidimensional risk patterns, necessitating the development of new prediction methodologies. Materials and Methods: This paper presents a structured predictive healthcare paradigm that use machine learning approaches to diagnose T2DM early on. A benchmark clinical dataset was used to compare the performance of Decision Tree, Random Forest, Support Vector Machine, and Artificial Neural Network models. A hybrid ensemble model based on a stacking method was created to combine the capabilities of various algorithms. Accuracy, precision, recall, F1-score, and ROC-AUC were used to measure model performance, while robustness was tested using 10-fold cross-validation. Model interpretability was integrated using SHAP and LIME to assess feature significance and facilitate clinical transparency. Results: The hybrid ensemble model outperformed particular models by 2-4%, achieving 87.3% ± 1.2 accuracy and ROC-AUC of 0.91 ± 0.01. The model showed less variability over cross-validation folds, suggesting better stability and generalizability. Feature relevance analysis revealed that plasma glucose concentration, body mass index, and age were the most important predictors, which is consistent with published clinical findings. Conclusion: The suggested hybrid ensemble architecture improves predictive accuracy, resilience, and interpretability for diabetes prediction. The approach promotes early risk stratification by tying model outputs to clinically relevant risk indicators and has the potential to be integrated into decision-support systems. Future study should focus on validation using real-world datasets and deployment in clinical settings.

Keywords : T2DM; Machine Learning; Predictive Healthcare; Hybrid Ensemble Model; Early Disease Detection; SHAP; LIME; Clinical Decision Support; Data-Driven Medicine; Risk Stratification