<p>Chronic Kidney Disease (CKD) is a progressive condition that should be detected early and accurately as it can improve patient outcomes. The objective of this study is to propose an improved XGBoost-based CKD detection framework that has higher classification performance by optimizing using Improved Brainstorm Optimization (BSO). Proposed approach is a comprehensive preprocessing pipeline consisting of mode imputation, Min Max normalization, Principal Component Analysis PCA for dimension reduction. Improved BSO is used to perform hyperparameter tuning for XGBoost key parameters, balancing exploration and exploitation dynamically to optimize XGBoost’s key parameters. Fivefold cross validation is used to evaluate the model, to make it robust and generalizable. Experimental results on the UCI CKD dataset show that the XGBoost and Improved BSO models achieve an accuracy of 97.5%, surpassing several baseline models, including Logistic Regression, CatBoost, and Stochastic Gradient Boosting. Class weighting in XGBoost helps to deal with class imbalance, PCA helps to remove redundant features, thereby improving computational efficiency. Stability of the presented model is confirmed with standard deviation analysis across cross validation folds. The results show that Improved BSO performs well in hyperparameter tuning and achieves a very accurate and efficient CKD detection model. Deep learning-based models and federated learning are explored for further improving CKD diagnosis in real world clinical settings in future work.</p>