Comparison of Naive Bayes Classifier with Feature Selection Gain Ratio on Data Classification

Abstract

In this study, the authors propose a process of increasing accuracy in Naïve Bayes with a combination of feature selection using the gain ratio and Relief-F methods. The cause of the less than optimal accuracy in Naïve Bayes compared to other classification methods is due to the less significant influence of features and the relatively low percentage of influence of data in determining the class of new data. The Gain Ratio and Relief-F methods are used to select features that have a poor correlation with the data being tested. The test of the proposed method is to compare the accuracy obtained from the Naïve Bayes method without using feature selection with Naïve Bayes using Gain Ratio and Relief-F feature selection. The test results obtained were the proposed Gain Ratio and Relief-F methods, the gain ratio method did not increase while the Relief-F method was able to increase the classification accuracy of naïve Bayes with an increase obtained of 0.2928% when compared to the Naïve Bayes test without feature selection.