A Systematic Review of Hybrid Intelligent Models for Early Detection of Diabetes Using Electronic Health Records

Abstract

Diabetes is a major global health concern due to its rising prevalence and the significant number of undiagnosed cases. Early detection using Electronic Health Records (EHRs) has become a key research focus, supported by advances in machine learning and artificial intelligence. This study presents a systematic review of hybrid intelligent models for early diabetes detection using EHR data from 2015 to 2026. The review analyzes various hybrid approaches, including ensemble learning, deep learning, Fuzzy logic-based, feature selection techniques, and optimization-based models. It also examines commonly used datasets, feature extraction methods, and evaluation metrics such as accuracy, precision, recall, F1-score, and AUC-ROC. Findings indicate that hybrid intelligent models generally outperform traditional machine learning methods by improving predictive accuracy and capturing complex nonlinear relationships in clinical data. However, issues such as missing data, lack of standardization, interpretability, and limited external validation remain present. The study highlights the need for explainable AI, federated learning, and multimodal data integration to improve clinical applicability. Overall, this review provides insights into current methodologies and identifies future directions for developing more robust, scalable, and clinically applicable diabetes prediction systems.