A Novel Approach for Diabetes Mellitus Detection Using a Modified Binary Multi- Neighbourhood Artificial Bee Colony Algorithm with Mahalanobis-Based Feature Selection (MBMNABC-Ma) and an Optimized Decision Forest Framework

Abstract

Diabetes is a critical global health issue caused by high blood sugar (hyperglycemia), leading to complications like cardiovascular disease, blindness, neuropathy, and kidney failure. Machine learning (ML) algorithms improve both the accuracy and efficiency of medical diagnoses. This study applies a Modified Binary Multi-Neighbourhood Artificial Bee Colony with Mahalanobis- based (MBMNABC-Ma) for a feature selection algorithm, combined with diverse ML models for diabetes identification. Compared to the conventional Binary Multi-Neighbourhood Artificial Bee Colony (BMNABC), MBMNABC-Ma improves classification accuracy and reduces computational complexity. Five diabetes datasets were analyzed using a 70-30% holdout cross-validation. The MBMNABC- Ma model, trained on Optimal Decision Forest (ODF) with Random Forest Ensemble (RFE), demonstrated high effectiveness. It achieved 97.23% accuracy on the Merged Datasets (comprising 130 US and PIMA datasets), 97.93% on the Iranian Ministry of Health Dataset, 96.05% on the Questionnaire Dataset, 98.39% on the Hospital of Sylhet Dataset, and 80.98% on the PIMA Dataset, with high specificity and sensitivity scores across all cases.