An Intelligent Framework for Early Chronic Kidney Disease Prognosis

We propose a hybrid wrapper approach, “Statistical-Recursive Feature Elimination” (S-RFE), to obtain an optimal and highly predictive subset of features. Max Voting Ensemble (MVE) optimization is applied to effectively predict diseases with optimal features, where the problem of finding the best classifier is formulated by representing multiple objective functions as a single objective function. Significant disease predictors were validated using statistical significance values and AUC-ROC. The proposed model has been experimentally validated for real data, and a comparison of models with different benchmark classifiers is also presented. We have achieved maximum accuracy (99.25%) and F1-score (99.2), with accuracy (0.99), having the ideal count of predictive features. Our prototype generates alerts for patients at risk, as well as for any new patients who may be at risk.