An Explainable Machine Learning Model for Diagnosis of Chronic Kidney Disease Using Regular Health Informatics

People all over the globe are impacted by the serious public health problem known as chronic kidney disease (CKD). It is associated with unfavorable health outcomes that can occasionally result in renal failure and other chronic illnesses including cardiovascular disease. Early identification and appropriate treatment may be able to lessen the burden of CKD. In recent years, several clinical decision-making systems on CKD detection have been developed by various researchers to facilitate and confirm an efficient diagnosis. In this research, CKD advancement is assessed directly from a dataset comprising regular health check information for 1 million participants. We have utilized Chi-Square, Minimum Redundancy Maximum Relevance (mRMR), and Recursive Feature Elimination (RFE) feature selection algorithms to select the most crucial features. K-nearest Neighbors (KNN), Naive Bayes (NB), and Support Vector Machine (SVM) are three state-of-the-art classification algorithms used in this study. The findings have been calculated for each classifier utilizing the selected features from the feature selection algorithms. The following measures are used to assess the performance of the models: accuracy, sensitivity, specificity, F1-score, and Area Under the Curve (AUC). The performance of each classification method is impressive. The SVM method, however, has outperformed all other used algorithms and has achieved 99.29% accuracy. An AI-based model explanation method, SHapley Additive exPlanations (SHAP), has been used to disclose the importance of the features in classification. Among the top features contributing most to the prediction model, age, sex, HMG, smoking state, height, waist, drinking state, and ALT is cost-effective regular health information.