Diabetes, a persistent metabolic disorder, results in elevated blood sugar levels, contributing to long-term damage to the kidneys, blood vessels, eyes, heart, and nerves. Type 1 Diabetes, formerly insulin dependent or juvenile, results from minimal pancreatic insulin production, necessitating external insulin for management. Most of those affected by diabetes, approximately 422 million people globally according to the World Health Organization (WHO), reside in middle- and low-income countries, resulting in 1.5 million annual diabetes-related deaths. Experts in the medical field are actively researching diabetes prediction for improved accuracy in early detection and management of the condition. The main objective is to compare various classifiers and feature selection methods to enhance prediction accuracy, contributing to more effective diagnosis and treatment approaches in healthcare. The study evaluated Cubic KNN, Linear SVM, Subspace Discriminant Ensemble (SDE), and Kernel Naïve Bayes classifiers, applying PCA feature selection technique to enhance early detection of diabetes, offering valuable insights for improving predictive accuracy in healthcare. In the study, raw data underwent preprocessing and the PIMA Indians diabetes dataset was utilized for experiments. Results showed that the accuracy achieved for Cubic KNN was 86.7%, for Liner SVM 85.9%, for Subspace Discriminant Ensemble, and for Kernel Naïve Bayes 85.3%, indicating Cubic KNN as the most accurate classifier.