An Ensemble Feature Selection Method for Prediction of CKD

Abstract

Feature Selection is an important precursor to prediction and classification of medical data. Medical data mining is evolving at a faster rate and the current machine learning algorithms need to be imbibed with intelligent prediction and classification systems to handle the huge medical data. Since the huge volume of data stored in the medical database may be prone to ‘curse of dimensionality’, it becomes necessary to adopt methods that handle the issues of high dimensional data and improve the stability of the selected method. One such solution as envisaged in recent literature is the ensemble feature selection technique that combines the results of filter and wrapper methods to select the most discriminatory features for diagnosis of chronic diseases. In this paper, an ensemble feature selection technique is applied to Chronic Kidney Disease (CKD) dataset. Density based Feature Selection (DFS) method is used as a filter approach to rank the features of CKD. The results of DFS method is given to a wrapper based optimization technique named Improved Teacher Learner Based Optimization (ITLBO) algorithm to find the optimal feature set that contains the most important features for prediction of CKD with high accuracy. The results of ensemble feature selection method are evaluated using SVM, Gradient Boosting, and CNN classification algorithms. Experimental results reveal that the proposed method is able to achieve high classification accuracy of 93% for SVM, 97% for Gradient Boosting and 97.75% for CNN respectively for the derived optimal feature subset.