Early Detection of Chronic Kidney Disease Through Explainable Clinical Intelligence

Abstract

Chronic kidney disease (CKD) is a developing condition which frequently goes unrecognized in its early stages, resulting in delayed treatment and poor clinical outcomes. Early detection is, therefore, fundamental to slow down the disease sequence and increase patient care. An explainable clinical intelligence framework was designed for early detection of CKD by combining machine learning (ML) algorithms and explainable artificial intelligence (XAI). A publicly available dataset of 400 patient records with 24 routinely collected clinical variables was used for a retrospective analysis. Following data preprocessing, five supervised machine learning models Logistic Regression, Random Forest, XGBoost, LightGBM, and CatBoost were created and tested with an 80:20 stratified train-test split. The performance of the model was evaluated using accuracy, precision, recall, F1-score and receiver operating characteristic area under the curve (ROC-AUC). SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) were used to justify the prediction results and discover the important clinical features. Ensemble learning models showed better predictive performance in which the highest classification accuracy was achieved by the models Random Forest, XGBoost, and CatBoost. Serum creatinine, haemoglobin, albumin, blood urea, and specific gravity were always found to be the most influential predictors of CKD in the explainability analysis. The results of these studies show that the fusion of effective machine learning models with explainable AI offers a reliable, transparent, and clinically significant approach to early detection of CKD.