Prediction of chronic kidney disease using different classification algorithms

摘要

Chronical kidney disease (CKD) is a common kidney function problem that causes deterioration of kidney performance and leads to kidney failure. An early diagnostic methodology to determine kidney functionality is essential and extremely important in many cases. In this study, different classifiers were applied for the classification of a CKD dataset. The algorithms were applied using random tree, decision table (DT), K-nearest neighbor (K-NN), J48, stochastic gradient descent (SGD) and Na?ve Bayes classifiers, and a prediction model was proposed based on feature selection to efficiently predict CKD cases. Results showed that the J48 and decision table classifiers outperformed the other classifiers with accuracies of 99%, ROCs equal to 0.999 and 0.992, MAEs of 0.0225 and 0.1815, and RMSEs of 0.0807 and 0.2507, respectively. A sensitivity analysis of selected classifiers was implemented to evaluate the performance of these classifiers with changes in their parameters. The J48 and decision table classifiers outperformed all other classifiers with an accuracy of 99% and RMSEs of 0.0807 and 0.2507, respectively. Additionally, the results showed an enhanced classification performance for K-NN (K?=?1). Na?ve Bayes and decision table classification were enhanced to 99.75%, 98.25% and 99.25%, respectively, when feature selection methods were applied, and only a handful of features were used for classification of the CKD dataset, in which such an enhancement can add value and support healthcare provided to identify certain CKD cases at early stages using the presented selected features.