Explainable AI machine learning framework for chronic kidney disease prediction utilizing electronic health records.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Medical Informatics

Background:

• Chronic Kidney Disease (CKD) poses a significant global health burden, often undetected until advanced stages.
• Early diagnosis and intervention are critical for improving patient outcomes and reducing morbidity/mortality.
• Existing diagnostic methods may lack the precision for timely detection.

Purpose of the Study:

• To develop and validate an explainable artificial intelligence (AI) driven machine learning (ML) framework for predicting Chronic Kidney Disease (CKD).
• To identify key clinical variables contributing to CKD prediction using electronic health records.
• To enhance the interpretability of ML models for clinical application in early CKD detection.

Main Methods:

• Utilized a dataset of 398 patients from Pakistan Kidney Center, analyzing various ML models including Random Forest (RF), Gradient Boosting, and XGBoost.
• Employed feature selection techniques (mutual information, DT, RF) to identify crucial clinical variables for prediction.
• Evaluated model performance using stratified cross-validation and statistical tests, with interpretability provided by SHAP and LIME.

Main Results:

• The Random Forest (RF) model achieved a high accuracy of 98% in CKD prediction.
• Key biomarkers such as estimated glomerular filtration rate, creatinine, and urea were identified as strong predictors.
• Explainable AI methods (SHAP, LIME) confirmed the clinical relevance of identified features, providing transparent insights.

Conclusions:

• The proposed explainable AI framework offers a robust, reliable, and interpretable solution for early CKD detection.
• The study highlights the effectiveness of selected clinical features in driving predictive performance across multiple ML models.
• This approach provides a practical and generalizable tool for clinical settings to improve CKD diagnosis.