Email Feature Classification and Analysis of Phishing Email Detection Using Machine Learning Techniques

Abstract

Due to a lack of security measures and user knowledge, email users are increasingly becoming the targets of this inescapable harassment as the number of phishing emails increases. Phishing emails have a significant impact on email systems. They attempt to steal users’ information and damage the email client's reputation. The most phishing emails have ever been sent, according to the APWG (Anti-phishing working group), were detected in the third quarter of 2022, when 1,270,883 total phishing attacks were recorded. Previous researchers had proposed methods of phishing email filtering that achieved accurate results on prefetched datasets, and email clients such as Gmail, Yahoo and Outlook have also developed filtering systems that filter phishing emails. Despite these contributions, it is still possible to get phishing emails daily because of their variety, ambiguity, and evolution. This paper introduced new features that were identified in modern emails and conducted two types of experiments using machine learning techniques. The first experiment focused on how the extracted features weighted the importance when classifying legitimate advertisements and phishing emails. And the second experiment focused on legitimate emails and phishing emails. We also evaluated these machine learning algorithms on a real dataset that we gathered, and some of the results were below par when compare with prefetched datasets.

Topics & Concepts