Hybrid Fuzzy Logic and Logistic Regression Model with Recursive Feature Elimination for Enhanced Prediction and Clinical Decision Support in Type 2 Diabetes Mellitus Among Adults Aged 35 to 45

This paper presents a hybrid model combining fuzzy logic, recursive feature elimination (RFE), and logistic regression to predict type 2 diabetes mellitus (T2DM) in adults aged 35–45, addressing uncertainty in clinical data, overfitting, and interpretability challenges. Fuzzy logic models uncertainty in variables like fasting plasma glucose (FPG), HbA1c, and BMI using optimized triangular membership functions. RFE selects optimal features, and logistic regression with enhanced regularization mitigates overfitting to provide interpretable probabilities. Evaluated on five datasets—Pima Indians Diabetes Database (PIDD), NHANES, Iraqi Patient Dataset of Diabetes (IPDD), UK Biobank, and CTGAN synthetic data—the model achieves up to 8% higher AUC (0.92 ± 0.02, p<0.01 vs. Random Forest) and outperforms Neural Networks, Transformers, and CNNs in accuracy (89.7 ± 1.0% on IPDD), sensitivity (86.5 ± 1.5% on NHANES), and specificity. A decision tree provides actionable clinical guidance, enhancing early T2DM detection and costeffective interventions.