Accuracy Prediction of Classification and Forecast using WEKA Tool by Example: Chronic Kidney Disease

Abstract

Introduction: Renal failure due to kidney disease can be avoided with early diagnosis. Disease markers able to anticipate renal failure at an asymptomatic stage and thus, the onset of chronic kidney disease in a human subject can be predicted using, for example, data mining techniques. The present study focuses on building a decision tree and predicting the accuracy of machine learning classifiers to forecast kidney disease using the CKD dataset. Methods: The dataset in the current study includes information from 400 samples (instances) and 25 attributes retrieved from the freely available UCI machine learning repository. The accuracy of prediction of classifiers was conducted with the WEKA software tool using 14 algorithms. The performance evaluation of the models was done with accuracy, precision, recall and F-measure. Results: The lowest performance was given by Stacking and Vote classifiers. Moderate performance evaluation was observed for Logistic, Naïve Bayes, Random Tree, and Voted Perceptron. The best performances were observed for Random Forest, Multilayer Perceptron, Logit Boost, J48, Decision Table, Bagging, PART, and SMO. The following two were statistically significant: Random Forest and Multilayer Perceptron . Conclusion: The decision tree could successively depict the contribution of serum creatine, pedal edema, diabetes, hemoglobin, and specific gravity of blood in tracing the prevalence of CKD in a prospective patient.