KMAN Publication Inc. (KMANPUB) Health Nexus In Press 2027 01 01 A Diabetes Diagnosis Approach by Feature Selection Using Group Learning Optimization (GLO) Algorithm and Apache Spark Distributed Processing Technology A Diabetes Diagnosis Approach by Feature Selection Using Group Learning Optimization (GLO) Algorithm and Apache Spark Distributed Processing Technology 1 24 EN Journal Article 2026 05 21 <p>This study aimed to develop and evaluate an accurate and computationally efficient diabetes diagnosis framework integrating GAN-based class balancing, binary Group Learning Optimization (GLO) feature selection, and Apache Spark distributed machine-learning classification. The study used the Pima Indians Diabetes Dataset containing 768 records with eight diagnostic features and a binary diabetes outcome. Data were normalized to the [0,1] range, and a Generative Adversarial Network was used to balance minority and majority classes. A binary version of the GLO algorithm was developed for feature selection by minimizing a fitness function combining classification error and the number of selected features. The dataset was divided into 70% training and 30% testing subsets. A multilayer perceptron was used to evaluate candidate feature vectors during optimization. The reduced-dimensional data were subsequently processed within an Apache Spark master-worker architecture and classified using Decision Tree, Random Forest, Support Vector Machine, and majority voting. GLO was additionally evaluated against TLBO, WOA, SHO, WSA, and HHO using benchmark optimization functions. GLO achieved the best mean optimization rank of 1.86 and the lowest local-optimum entrapment rate of 7.5%. In feature-selection experiments, GLO achieved 89.76% accuracy, 88.60% sensitivity, and 88.41% precision, outperforming CSA, GWO, GOA, PSO, and GA in diagnostic accuracy. GAN balancing improved Decision Tree, Random Forest, and SVM accuracy to 93.21%, 94.62%, and 91.63%, respectively, while majority voting achieved 93.92% accuracy, 93.67% sensitivity, and 92.84% precision. In the distributed Spark implementation, Random Forest achieved the highest diagnostic performance, with 98.97% accuracy, 98.64% sensitivity, and 98.62% precision. Computational acceleration increased with the number of Spark clusters, with Decision Tree attaining a maximum speedup of 12.56. Integrating GAN-based balancing, binary GLO feature selection, and Apache Spark distributed processing provides an effective framework for improving diabetes classification accuracy while reducing dimensionality and computational burden.</p> https://www.journals.kmanpub.com/index.php/Health-Nexus/article/download/6008/11677