Leveraging Stacked Machine Learning Models to Advance Diagnostic Precision and Predictive Insights in Chronic Kidney Disease

Abstract

A considerable number of people worldwide suffer from chronic kidney disease (CKD), a progressive illness that often goes undiagnosed until it has advanced to a severe degree. Timely action and better patient outcomes rely on early diagnosis. However, traditional diagnostic methods are time-consuming and may lack consistency, especially in resource-constrained settings. To address this, our study explores the application of machine learning (ML) for early-stage CKD prediction. We used the publicly available CKD dataset from the UCI Machine Learning Repository, which includes 400 patient records across 14 clinical features. After performing thorough preprocessing—including handling missing values and converting categorical data—we applied multiple ML classifiers: Naïve Bayes, Decision Tree, Gradient Boosting, AdaBoost, and XGBoost. Each model was evaluated using 10-fold cross-validation to ensure reliability. The core of our approach lies in a stacking ensemble model, which combines predictions from three base learners—Naïve Bayes, Decision Tree, and Gradient Boosting—and passes them to a meta-learner based on AdaBoost. This layered learning framework was specifically designed to mitigate overfitting, which is evident in some individual models during evaluation. The proposed stacking model demonstrated the best performance among all tested models, achieving a training accuracy of 99.2% and a testing accuracy of 98.93%. Furthermore, it yielded a precision of 98.70%, a recall of 98.10%, and an F1-score of 98.40%, outperforming other standalone algorithms in accuracy and stability.