Zero-Day attack prevention Email Filter using Advanced Machine Learning

Abstract

Preventing email spams continues to be a challenge as the attackers are using new techniques that circumvent the existing spam filters. Therefore, a smart email filter that can identify zero day attacks is necessary. In this paper, we propose an approach which not only looks at the text of the body of the email but also handles the embedded phishing URLs and attached spam images. The proposed approach uses several advanced Machine Learning algorithms to classify the emails and provides a structured process to identify the spams. We use lazyPredict library for selecting the best performing machine learning models. Our case studies using standard data sets show that these smart filters perform well in identifying spams and preventing zero-day attacks. Our analysis of results shows that Stacking classifier performs better with accuracy score of 0.97 for phishing URLs detection. Whereas, the perceptron classifier with accuracy of 0.97 the top performer in detecting email spams. The performances of other algorithms are also reported.