Machine Learning Techniques in Chronic Kidney Diseases: A Comparative Study of Classification Model Performance.

Area of Science:

• Nephrology
• Medical Informatics
• Machine Learning

Background:

• Chronic kidney disease (CKD) prevalence is rising across age groups.
• Accurate CKD assessment and monitoring are vital for preventing kidney damage.
• Machine learning (ML) offers potential for rapid and precise disease detection in healthcare.

Purpose of the Study:

• To develop and evaluate an ML system for supporting CKD diagnosis.
• To compare the performance of six distinct ML algorithms for CKD detection.

Main Methods:

• Utilized the UCL machine learning database for CKD data.
• Processed data by imputing missing values and applying polynomial techniques for feature enhancement.
• Employed feature-based stratified splitting with K-means clustering.
• Implemented and compared Random Forest, SVM, Naive Bayes, Logistic Regression, KNN, and XGBoost algorithms.

Main Results:

• Random Forest, XGBoost, SVM, and Logistic Regression achieved 100% accuracy.
• Naive Bayes reached 97% accuracy, and KNN achieved 93% accuracy.
• The developed ML system demonstrated high efficacy in CKD diagnosis.

Conclusions:

• ML models, particularly Random Forest, XGBoost, SVM, and Logistic Regression, show exceptional promise for accurate CKD diagnosis.
• The study highlights the effectiveness of ML in enhancing early detection and management of CKD.
• Further integration of such systems can significantly aid clinicians in patient care.