Prediction of Chronic Kidney Disease Based on Comparison of Machine Learning Models

Abstract

Chronic kidney disease (CKD) has become an important issue affecting global public health security due to its complex and variable pathogenic factors. Machine learning models with good predictive performance for CKD can break the limitations of traditional diagnostic methods and effectively control the burden of CKD on patients and the harm to human health security. In this paper, eXtreme Gradient Boost (XGBoost), logistic regression, and Support Vector Machine (SVM) were adopted for the prediction training of CKD dataset. This research comprehensively evaluated the model performance through accuracy rate, precision, recall rate, F1 value, AUC value, ROC curve, and scatter plot based on the T-SNE algorithm. Finally, the research concluded that XGBoost had the best performance. Subsequently, the research statistically analyzed features that were more important for the research through the plot_importance function and plotted a horizontal bar chart of the top 10 features in terms of importance. This research can help improve the efficiency of relevant practitioners and researchers in diagnosing CKD, contribute to reducing the burden on patients and enhancing the control of the incidence of CKD, and make contributions to public health issues.