Entropy-Based Feature Selection Classification Approach for Detecting Phishing Websites

Abstract

In this modern era of computing, phishing is a noxious cyber-attack in which the user’s sensitive personal or financial information is obtained by an illegitimate website. The detection of phishing websites is very important to cope with phishing incidents to protect the user’s credentials on the web. The uniform resource locators (URLs) of phishing websites look almost like the URLs of legitimate ones however differ in several aspects. In this research work, we have presented an entropy-based feature selection (FS) approach with machine learning (ML) classification to differentiate phishing website patterns from the legitimate ones. The main aim of this study is to identify the set of most dominant features associated with the phishing website patterns. This research work is divided into two phases. In the first phase, we have proposed different FS methods to mine the optimal feature subset having the most dominant features representing the phishing patterns. In the second phase, we have implemented five different classification techniques namely k-nearest neighbors (KNN), support vector machine (SVM), decision tree (DT), random forest (RF), and multi-layer perceptron (MLP) classifier to identify the insecure patterns on the web and classify them as phishing websites by making use of optimal feature subset. Finally, the performance of ML classifiers is evaluated using different performance evaluation metrics i.e., specificity, sensitivity, precision, accuracy, and ROC-AUC (receiver operating characteristic - area under the ROC curve) curve. The results of the paper proved that the integration of the entropy-based FS method with the ML classifiers helps to improve their efficiency and accuracy. Furthermore, the comparison of the results indicates that MLP classifiers with entropy-based FS outperformed other classifiers for detecting phishing websites and making the web more secure for the people to surf.