FEATURE SELECTION METHODS FOR CLASSIFICATION – A COMPARISON

Kavitha C.R, Mahalekshmi T

Abstract: Feature selection refers to the process of choosing the most significant features for a given task, while discarding the noisy, irrelevant and redundant features of the data set. These noisy feature set might mislead the classifier. Feature selection technique reduces the dimensionality of the feature set of the data set. The main aim of this work is to perform binary and multiclass classification more accurately using reduced number of attributes. This paper proposes two different feature selection methods. The first feature selection method is done using Information Gain and Forward Selection (IGfwS). The second feature selection is performed using Recursive Feature Elimination with SVM (SVMRFE). Then rough set theory was applied to both the feature selection methods to obtain hybrid feature selection methods RST+SVMRFE and RST+ IGfwS. Further, a comparative study of all the four feature selection methods was performed. From the results of the study, it is found that the feature selection is a very important data mining technique which helps to achieve the good classification accuracy with the reduced number of attributes. Based on the comparative analysis conducted the feature selection methods SVMRFE and RST+SVMRFE shows better performance than other feature selection methods considered under the study. And the random forest classifier achieves the maximum accuracy with all the datasets on which SVMRFE and RST+SVMRFE feature selection methods were applied.

Keywords: Feature Selection, Classification, Rough Set, Information Gain, Ranking,

DOI: https://doi.org/10.15623/ijret.2017.0606022