<p>Chronic kidney disease (CKD) is a significant global health challenge, leading to high morbidity, mortality, and healthcare costs. Early detection and accurate staging of CKD are crucial for effective management and improved patient outcomes. While most previous studies have used machine learning techniques to detect CKD at an early stage, they have not primarily focused on predicting specific stages of the disease. In this study, we explored both binary classification (presence or absence of CKD) and multi-class classification (prediction of CKD stages) to provide more actionable insights for healthcare professionals. We employed Random Forest, Decision Tree, and XGBoost classifiers in combination with feature selection methods, including Mutual Information, Genetic Algorithm, and Bat Algorithm, as well as the Synthetic Minority Over-sampling Technique (SMOTE) to enhance model performance and address class imbalance. Additionally, we utilized K-Nearest Neighbors (KNN) imputation to manage missing values. To ensure transparency and interpretability of the predictions, we integrated the explainable AI method SHAP (SHapley Additive exPlanations), which provides valuable insights into how the model features influence CKD predictions. The experimental results showed that in binary classification, Random Forest with GA-selected features and XGBoost with MI- or GA-selected features achieved the highest accuracy of 99.17%. For multi-class classification, the XGBoost classifier with GA-selected features achieved the highest accuracy of 92.00%. Overall, our approach provides an effective framework for CKD prediction and staging, combining advanced machine learning techniques, feature selection, and explainable AI methods to enhance model accuracy and interpretability, ultimately contributing to improved patient outcomes.</p>