Development of an ensemble approach to chronic kidney disease diagnosis

Abstract

Chronic kidney disease is a major health crisis globally killing millions of people every year emanating as a result of unhealthy lifestyle and hereditary factors. The need for prompt and accurate diagnosis triggered the application of data mining techniques. In recent times, data mining techniques have been subjected to extensive research in chronic kidney disease diagnosis with emphasis majorly on accuracy either through the simplification of the disease by performing feature selection in addition to preprocessing or not before classification. This paper employs two ensembles approaches namely: Bagging and Random Subspace methods on three base-learners – k Nearest Neighbours, Naïve Bayes and Decision Tree for improving the classification performance of the models. Prior to classification, data preprocessing were carried out to handle missing values and data scaling to normalize the range of independent variables. The performances of the model were evaluated with accuracy, specificity, sensitivity, kappa and ROC criteria. Empirical results on Chronic kidney disease dataset obtained from UCI machine learning repository shows that the ensemble techniques provides better result than individual base learners performances with random subspace having the best performance than bagging in most of the considered situations.100% accuracy of prediction is attained using random subspace ensemble on KNN classifier. Hence the model is suitable for efficient diagnosis of chronic kidney disease.

Keywords

• Chronic kidney disease (CKD)
• Data mining
• Bagging
• Random subspace Naive Bayes
• KNN
• Decision tree