Prediction of Chronic Kidney Disease with Machine Learning Models and Feature Analysis Using SHAP

Abstract

The world is significantly impacted by chronic kidney disease (CKD), both in terms of the health and financial costs. CKD is becoming a bigger issue globally, especially in low- and middle-income nations. According to the Global Burden of Disease Survey, 697.5 million people worldwide suffered from chronic kidney disease (CKD) in 2019.Addressing the burden of CKD requires a comprehensive approach that includes prevention, early-detection, and effective management of the condition. The main objective of this research work is to utilize Machine Learning methodologies to facilitate the diagnosis of Chronic Kidney Disease (CKD) by leveraging relevant clinical details. To accomplish this, the classification models Logistic Regression, Random Forest, Voting Classifier and Support Vector Machine are employed to distinguish patients with CKD from those without. The evaluation shows that according to the evaluations matrices Voting Classifier with soft voting showed an average classification accuracy of 98%, f1-score of 97.4%, precision of 95%, recall of 100%. Random Forest classifier showed an average classification accuracy of 96%, f1-score of 95%, precision of 90.4%, recall of 100%. Logistic regression classifier showed an average classification accuracy of 94%, f1-score of 92.6%, precision of 86.3%, recall of 100%. Support Vector Machine classifier showed an average classification accuracy of 90%, f1-score of 88.3%, precision of 79.1%, recall of 100% and proves that Voting Classifier performed well which is immediately followed by Random Forest and then Logistic Regression. Furthermore, the SHAP (SHapley Additive exPlanations) model interpretability technique is utilized to analyze the significance of each feature in determining output.