Explainable Machine Learning Framework for Chronic Kidney Disease Prediction Using Multi-Model Comparative Analysis and LIME Interpretability

Chronic Kidney Disease (CKD) is a painful and usually symptomless disorder, which requires early diagnosis to avoid renal failure. This study describes a predictive machine learning model of CKD, which combines several classification algorithms and model interpretability by Local Interpretable Model-Agnostic Explanations (LIME). A CKD dataset with 400 patient records and 26 clinical attributes, which is publicly available, was trained and tested using five supervised learning algorithms, including Logistic Regression (LR), Support Vector Machine (SVM), K-Nearest Neighbours (KNN), the Random Forest, and XGBoost. Preprocessing of the data involved normalisation, categorisation, and the filling of missing values to ensure consistency of the models. It was experimentally proven that ensemble models were higher in performance as XGBoost and the Random Forest had 98.8% and 98.5% accuracy$(\mathrm{F} 1=0.99$and$\text{AUC}=1.00)$respectively, which significantly surpassed all other classifiers. The interpretability offered by LIMF revealed the effects of the main biomarkers (serum creatinine, blood urea, haemoglobin, and specific gravity) on the prediction made by the model. The suggested framework produces accurate diagnostic results and, at the same time, improves clinical confidence with its interpretable predictions being transparent. This will be further expanded in future work, based on federated and cross-institutional datasets, to perform scalable CKD diagnosis without compromising privacy.