Comparative study of classification algorithms in chronic kidney disease

Abstract

Chronic Kidney Disease is a very dangerous health problem that has been spreading globally due to alterations in life style such as food habits, changes in the atmosphere, etc. So it is essential to decide on any remedy to avoid and to predict the disease in early stage which helps to avoid wastage of life. We show that feature selection approach is well suited for chronic kidney disease prediction. Principal Component Analysis is one of the feature selection techniques that filters out less important attributes; it also picks attributes of importance from the dataset. We also compare different data classification approaches in terms of how accurately they predict chronic kidney disease. We examine Decision stump, Rep tree, IBK, K-star, SGD and SMO classifiers using performance measures like Kappa statistics, Receiver Operating Characteristic, Mean Absolute Error and Root mean squared Error using WEKA. Accuracy measures used to compare classifiers are Recall, F-measure and Precision by implementing on WEKA. WEKA-a software for data mining, that uses collection of algorithm for data mining. It is possible to apply these algorithms directly to the data or call them from java code. Results obtained show better accuracy measures for Decision stump and Rep tree where the mean absolute error were less with error rate of 0.010 and 0.012 respectively.