Early identification of chronic kidney disease is crucial for initiating treatment promptly and lowering the risk of adverse outcomes. While machine-learning classifiers-including Random Forest, Gradient Boosting, Extra Trees, XGBoost, Naïve Bayes, Support Vector Machine, and Decision Tree-often achieve strong diagnostic performance, their limited interpretability can reduce clinician confidence and slow real-world adoption. To improve transparency, explainability methods such as SHAP and LIME are widely used to clarify how input features influence individual predictions, helping practitioners validate and rely on model outputs. At the same time, predictive performance depends heavily on hyperparameter selection; although GridSearchCV is commonly used, it may require substantial computation. In this work, we present a framework that combines RandomizedSearchCV for efficient hyperparameter optimization with SHAP and LIME to deliver both accurate and interpretable CKD prediction. Experimental results demonstrate an accuracy of 99%, over existing models. SHAP/LIME explanations elucidate the principal factors influencing each decision. By achieving a balance between performance and interpretability across various classifiers, the proposed approach enhances the reliability of clinical decision-making and augments the practicality of machine learning-based decision support in healthcare.