Improving diabetes diagnosis using metaheuristic-based ensemble classification method

Diabetes is a chronic metabolic disorder that, if left untreated, can lead to severe complications and reduced quality of life. Early diagnosis is essential for preventing its adverse effects; however, recent diagnostic methods face challenges such as data imbalance, missing values, and inadequate model accuracy. To address these issues, we propose a novel hybrid framework for diabetes diagnosis, integrating advanced data mining techniques. The proposed methodology combines the Borderline Synthetic Minority Oversampling Technique (BSMOTE) with the Giza Pyramids Construction (GPC) algorithm to generate synthetic samples and mitigate data imbalance. Missing data are imputed using the k-Nearest Neighbors (k-NN), ensuring dataset integrity. For classification, an ensemble approach is employed that integrates k-NN, Support Vector Machines (SVM), Random Forest, and Extreme Learning Machine (ELM) through a majority voting scheme.. The framework is evaluated on the PIMA Indian Diabetes Dataset (PIDD), achieving prediction accuracy ranging from 91% to 93.9%, representing up to a 6% improvement over existing methods. Key contributions include the effective handling of imbalanced and incomplete datasets and the introduction of a robust hybrid classification framework, which demonstrates significant potential for improving early diabetes diagnosis and management.