Comparison of feature selection method for chronic kidney data set using data mining classification analytical model

Abstract

Medical Data Analysis is one of the applications of data mining that plays a vital role in human's lives. Data mining has become famous in healthcare management to predict, to detect or to find the hidden patterns and information in health data. In the healthcare commerce different kind of data mining approaches are used to mine the interesting pattern of diseases using the statistical medical data with the help of different machine learning techniques. The conventional disease diagnosis system uses the observation and knowledge of doctor without using the complex clinical data. The planned system assists doctor to predict sickness properly and also the prediction makes patients and medical insurance suppliers benefited.    This paper implemented a feature model construction and comparative analysis for improving prediction accuracy of chronic kidney disease dataset in four phases. In first phase, Z-Score normalization algorithm is applied on the original kidney datasets collected from UCI repository. In the second phase of kidney dataset prediction, by the use of Step wise Regression Classification (SRC) Model and Built around the Random Forest Classification algorithm (BRFC) feature selection, subset (data) of Kidney dataset from whole normalized Kidney patient dataset is obtained which comprises only significant attributes. Third phase, classification algorithms are applied on the kidney data set. In the fourth phase, the accuracy will be calculated using Mean Absolute error (MAE), Root Mean square Error (RMSE), Relative Absolute Error(RAE), Root Relative Square Error(RRSE) and kappa values. MLP(Multilayer Perceptron Model)  and SVM (Support Vector Model) classification algorithm is considered as the better performance algorithm after applying BRFC and SRC feature selection. Finally, the assessment is done based on accuracy values