An Efficient Approach for Phishing Detection using Machine Learning

Abstract

The increasing number of phishing attacks is one of the major concerns of security researchers today. The traditional tools for identifying phishing websites use signature-based approaches which are not able to detect newly created phishing webpages. Thus, researchers are coming up with machine learning-based methods which are capable to detect and classify the phishing webpages with high accuracy if a large and variety of features are considered. However, building a classification model using a large number of features takes time which hampers the timely detection of phishing websites. Therefore, it is pertinent to shortlist a set of features using a feature selection method so that high-performance classification models can be developed in less time. In this chapter, we study the role of feature selection methods in detecting phishing webpages efficiently and effectively. A comparative analysis of machine learning algorithms is carried out on the basis of their performance without and with feature selection. Experiments are conducted on a phishing dataset with 30 features containing 4898 phishing and 6157 benign webpages. Several machine learning algorithms are used for obtaining the best results. Afterward, a feature selection method is employed to improve the efficiency of the models. The best accuracy is obtained by random forest both before and after feature selection with a significant improvement in model building time. The experiments demonstrate that employing a feature selection method along with machine learning algorithms can improve the build time of classification models for phishing detection without compromising their accuracy.