Chronic Kidney Disease (CKD) represents a significant global health challenge, with early detection being crucial for effective intervention. This study presents a comprehensive benchmarking analysis of ten machine learning algorithms for CKD prediction using a dataset of 1,659 patient records with 54 clinical, biological, and demographic variables. We evaluated traditional classifiers, ensemble methods, and neural networks through stratified 5-fold cross-validation, implementing three distinct feature selection strategies: genetic algorithm, recursive feature elimination (RFE), and feature importance-based selection. Our results demonstrate that ensemble methods consistently outperform traditional approaches, with XGBoost achieving 92.16% accuracy on the full feature set and up to 92.46% after RFE-based feature selection. CatBoost and Random Forest also showed superior performance, reaching 92.34% and 92.28% accuracy respectively with optimized feature subsets. The study reveals that proper feature selection significantly improves model performance while reducing complexity, with the most predictive features including serum creatinine, protein in urine, BUN levels, and fasting blood sugar. This comprehensive benchmarking provides valuable insights for healthcare practitioners and researchers in selecting optimal ML approaches for CKD prediction, particularly in resource-constrained settings where model efficiency and interpretability are crucial.