Hi-Le And Hitcle: Ensemble Learning Approaches For Early Diabetes Detection Using Deep Learning And Explainable Artificial Intelligence

Abstract

Early detection of diabetes is crucial for preventing severe complications and reducing long-term healthcare costs. With the increasing availability of medical data, artificial intelligence (AI) has emerged as a powerful tool for disease prediction. This study presents Hi-Le and HiTCLe, two novel ensemble learning frameworks for early diabetes detection that integrate deep learning models with explainable artificial intelligence (XAI) techniques. The proposed approaches combine the strengths of heterogeneous deep learners to improve predictive accuracy, robustness, and generalization across diverse patient datasets. While Hi-Le focuses on layered ensemble fusion, HiTCLe enhances decision-making through a hierarchical and task-correlated learning strategy. To address the black-box nature of deep learning, XAI methods such as feature attribution and local explanation models are incorporated to provide transparent and interpretable predictions for clinicians. Experimental evaluations conducted on benchmark diabetes datasets demonstrate that the proposed ensemble frameworks outperform traditional machine learning and standalone deep learning models in terms of accuracy, precision, recall, and F1-score. The integration of explainability further improves clinical trust and usability. Overall, Hi-Le and HiTCLe offer an effective, interpretable, and reliable solution for early diabetes diagnosis, supporting data-driven decision-making in modern healthcare systems.