Swarm Intelligence Feature Selection and Possibilistic Fuzzy c-Means Approach for Enhancement of Classifier Performance in Detection of Diabetes from Microarray gene

To combat the rising global burden of type II diabetes, early diagnosis is critical. This work proposes a two-stage machine learning approach for microarray data analysis. Stage 1 employs possibilistic fuzzy c-means (PFCM) for dimensionality reduction (DR). Stage 2 investigates Particle Swarm (PSOA) and Dragonfly (DFOA) optimization algorithms for feature selection (FS) and their impact on classification accuracy. Various machine learning algorithms were tested (NLR, LR, GMM, BLDA, SVM-RBF). Results show promising accuracy, with SVM-RBF achieving 91.42% without FS. However, the highest accuracy (92.85%) is achieved with DFOA-based FS for SVM-RBF. Interestingly, PSOA-based FS with SVM-RBF still yields a high accuracy (90%). These findings suggest DFOA outperforms PSOA and approaches without FS in enhancing the model's ability to detection of diabetes.