Chronic kidney disease (CKD) is a global health burden affecting millions, with early stages often remaining asymptomatic until irreversible complications like kidney failure or cardiovascular morbidity arise. Late detection significantly elevates mortality risks, with studies reporting up to 14.6% 1-year mortality in advanced CKD populations. Millions of people across the globe suffer from chronic kidney disease (CKD), which requires precise early detection tools. This study investigates the predictive performance of four K-Nearest Neighbors algorithm variants- Genetic algorithm-based (GAKNN), Optimal Fuzzy (OF-KNN), Mutual KNN, and Classic KNN on chronic kidney disease (CKD) datasets. The data used in this paper is taken from public sources such as Kaggle and UCI machine learning repository. The data was then pre-processed and the feature weights and parameters were optimized using genetic algorithm and bat optimization technique for GAKNN and OF-KNN models respectively. Performance of the variants of KNN algorithm is evaluated using metrics- precision, recall, accuracy, and F1-score. The results show that GAKNN algorithm consistently gives robust and stable performance due to low variability i.e. in the range of 0.005 to 0.017 on all the datasets. Hence, the analysis confirms that the GAKNN algorithm works better than other KNN variations in the detection of chronic kidney disease (CKD).