Optimizing Chronic Kidney Disease Prediction: A Machine Learning Approach with Minimal Diagnostic Predictors

Abstract

Chronic kidney disease (CKD) is a major public health issue that necessitates accurate diagnostic methods for effective management. This study involved training an open-source clinical dataset of 200 patients from Enam Medical College, comprising 28 clinical features, obtained from the UCI machine learning repository. After preprocessing to ensure a balanced dataset for objectivity, the data was split into training and testing sets in an 80:20 ratio. The research trained 22 machine learning models, including Naïve Bayes, decision trees, support vector machines (SVM), logistic regression, ensemble methods, kernel models, and neural networks. These models were evaluated using several metrics-accuracy, precision, recall, F1-score, and the area under the receiver operating characteristic (ROC) curve-computed through 5-fold cross-validation to assess their performance and ensure they were not overfitting or underfitting. The best-performing model was the Kernel Naïve Bayes, achieving a 96.55% accuracy, 95% precision, 98.28% recall, and 96.61% F1-score on the training dataset. For the test dataset, it showed a slight performance drop but remained robust with 92.86% accuracy, 87.50% precision, 100% recall, and 93.33% F1-score. Furthermore, feature selection techniques such as minimum-redundancy-maximum-relevance, Chi2, ANOVA, and Kruskal-Wallis tests were used to determine the most significant predictors. It was found that only four features-packed cell value, stages of glomerular filtration rate, specific gravity of urine, and albumin content in urine-were necessary for maintaining similar model performance. This systematic approach not only highlighted critical clinical features but also helped in simplifying the model complexity, which could benefit broader medical applications like lung cancer screening by reducing screen time, resources, and medical costs.