Interpretable Machine Learning for Chronic Kidney Disease Prediction: A SHAP and Genetic Algorithm-Based Approach

Chronic kidney disease (CKD) is a progressive condition that gradually diminishes kidney function, significantly contributing to global mortality rates. Developing a computer-aided diagnostic (CAD) system for efficient and accurate CKD detection is vital to enhancing diagnosis and treatment. This study aims to develop an interpretable CAD system that not only achieves high prediction accuracy but also provides insights into the factors influencing these predictions, aiding healthcare professionals in decision-making. We applied random forest (RF), decision tree (DT), logistic regression (LR), and XGBoost (XGB) classifiers to a UCI CKD dataset containing 400 samples and 24 attributes, addressing missing values with K-nearest neighbors (KNN) imputation. To enhance model performance, we employed a Genetic Algorithm for feature selection and utilized SHapley Additive exPlanations (SHAP) for interpretability. The classifiers achieved accuracies of 98.33%, 95.83%, 98.33%, and 99.17% for RF, DT, LR, and XGB, respectively, using all features. Feature selection with the Genetic Algorithm improved accuracy to 99.17% for both RF and LR, while DT accuracy increased to 97.50%, and XGBoost’s accuracy remained at 99.17%. SHAP analysis identified serum creatinine, hemoglobin, specific gravity, and albumin as the most influential features in CKD prediction. These findings demonstrate that the developed CAD system meets the study’s aims by offering high predictive accuracy alongside critical interpretability, enabling informed clinical decisions.