Authors:
Giacomo Giorgi
;
Andrea Saracino
and
Fabio Martinelli
Affiliation:
Informatics and Telematics Institute (IIT) of National Research Council, via G. Moruzzi 1, Pisa, Italy
Keyword(s):
Machine Learning, Deep Learning, Privacy-preserving, Email Authorship, Email Spoofing, Spear Phishing.
Abstract:
This paper proposes a novel email author verification aimed at tackling email spoofing attacks. The proposed approach exploits an authorship technique based on the analysis of the author’s writing style. The problem has been studied under two viewpoints, i.e. the typical sender verification viewpoint, already exploited in previous works, and the sender-receiver interaction verification, which to the best of our knowledge is a novel approach. Hence, we introduced the concept of end-to-end email authorship verification, which is focused on the analysis of the sender-receiver interactions. The proposed method implements a binary classification exploiting both standard machine learning classifiers based on the well-known text stylometric features and deep learning classifiers based on the automatic feature extraction phase. We have used a well-known email dataset, i.e. the Enron dataset to benchmark our approach, with the experiments showing an authorship verification accuracy reaching 9
9% and 93% respectively for the sender and the end to end verification scenarios. The proposed method has been implemented as an end-user support system in the Android environment for email spoofing attack detection.
(More)