A SHAP and LIME based Explainable AI Solution for Predicting Chronic Kidney Diseases

Abstract

Chronic Kidney Disease (CKD) presents a major global health issue, contributing to renal failure, cardiovascular problems, and elevated mortality rates. This research focuses on creating an effective machine learning (ML) model for CKD prediction, utilizing 25 features that represent different health indicators. We implemented three main algorithms: Logistic Regression (LR), K-Nearest Neighbors (KNN), and Decision Tree, along with extensive preprocessing, feature selection, and hyperparameter optimization. Based on accuracy, the models were evaluated, along with the confusion matrix, and ROC curves. Furthermore, we employed SHAP (SHapley Additive exPlanations) for model interpretability, offering insights via summary plots, waterfall plots, force plots, and dependence plots. Our results indicate high prediction accuracy, with a 10% increase in performance with the Decision Tree model achieving near-perfect performance, highlighting its potential for early CKD detection and contributing to timely medical interventions.