Leveraging Minimum Relative Entropy and Meta-Heuristic Feature Selection for Enhanced Microarray Gene Data Classification in Diabetes

Type II diabetes detection is a growing global health concern requiring early diagnosis. This study explores a two-stage method using microarray gene expression data. The first stage utilizes Minimum Relative Entropy (MRE) for dimensionality reduction. The second stage investigates the impact of Harmonic Search (HS) and Elephant Herding Optimization (EHO) for feature selection (FS) on classification performance. Various machine learning classifiers (NLR, LoR, GMM, SVM with linear, polynomial, and RBF kernels) were employed. Results show the proposed method, particularly SVM(RBF) without FS, achieves an accuracy of 91.42 percent. However, the highest accuracy of 95.71 percent is achieved with SVM(RBF) combined with FS techniques. This suggests MRE-based classification, especially with FS, improves Type II diabetic detection compared to approaches without feature selection.