Phishing Webpage Detection using Feature Selection Methods

Abstract

Phishing attacks are rapidly taking place around the globe. This makes it vital to have efficient phishing detection methods in place. All the datasets that are available are voluminous generally with a vast number of features. Furthermore, many of the features present are redundant or irrelevant and don’t substantially help in determining the final outcome. Therefore, it is necessary to identify those features and eliminate them to help reduce resources & time.This paper proposes two phishing detection techniques wherein one method incorporates ensemble feature reduction method and the other incorporates a feature reduction method based on average weight which help in eliminating irrelevant features and making a compact subset of the features to identify phishing attacks. These two methods are based on correlation, chi square, gain ratio, and information gain. The system uses Random Forest classifier which outperforms the rest of the classifiers. The comparison between both the methods is provided and the best method is determined taking factors like accuracy and computational time into consideration. The Phishing Webpage dataset is taken from Mendeley data.