A Novel Hybrid Machine and Deep Learning Model for Detecting Arabic Phishing Emails

Abstract

Phishing emails are becoming an increasingly popular type of cybercrime on the internet, affecting both businesses and individuals. The attackers generally use various methods to trick victims and extract personal information from them, such as bank details, home addresses, and account information. Many attempts have been proposed to tackle this issue by using filtering mechanisms or automated classification methods which require human intervention. However, this issue still remains a significant challenge. Additionally, attackers previously used manual methods to write phishing emails, however, recent AI tools have been used in means to generate such phishing emails. Therefore, this study aims to propose a hybrid machine and deep learning model to distinguish between content based phishing emails to categorize such emails into either real or fake, particularly, of the Arabic language. This model consists of BiLSTM (Bidirectional Long Short-Term Memory), GRU (Gated Recurrent Unit), and RF (Random Forest). In addition to this, a novel Arabic phishing email dataset has been developed. This imbalanced dataset consists of 418 phishing emails. Several experiments have been conducted in order to evaluate the proposed model. In addition, the sentence structure has been considered through the use of N-gram methods. Moreover, the experimental results show that the proposed model outperforms traditional machine learning classifiers and deep learning models. The model’s performance achieved an accuracy of 98.81%.