TY - GEN
T1 - Email Phishing Attack Detection using Recurrent and Feed-forward Neural Networks
AU - Lobo, Royce
AU - Abbas, Muhammad Naveed
AU - Asghar, Mamoona Naveed
N1 - Publisher Copyright:
© 2023 IEEE.
PY - 2023
Y1 - 2023
N2 - Email communication is significant in terms of its lateral and vertical information-sharing capability. The vulnerability of emails to phishing attacks poses significant threats to individuals and organizations, leading to financial losses, data breaches, and compromised security. This research study critically analyses features of email phishing through deep learning models. It examines the characteristics and patterns of phishing emails, such as misleading URLs, suspicious attachments, and linguistic cues that aim to manipulate the recipients. Then, the paper investigates the application of Artificial Neural Networks (ANNs) for analysing and detecting email phishing attacks. For experimentation, three ANNs, i.e. two recurrent, Long Short-Term Memory (LSTM) & Recurrent Neural Network (RNN), and one feed-forward, Convolutional Neural Network (CNN) are implemented and evaluated. The models are trained on the widely known Enron dataset of email samples, encompassing both spam and ham emails. The results show these ANNs outperform the traditional machine learning approaches as applied by contemporary research. While analyzing intra-model comparison using metrics, Recall & Type II Error, it is seen that CNN performs better than LSTM & RNN with 99% & 0.20% as compared to 31.32% & 38.60%, and, 52% & 27% respectively. Hence, the comparative analysis shows that the feed-forward (CNN) model performs better than recurrent (LSTM and RNN) models for detecting email phishing attacks. This finding can provide guidance to the research in this particular domain of email phishing.
AB - Email communication is significant in terms of its lateral and vertical information-sharing capability. The vulnerability of emails to phishing attacks poses significant threats to individuals and organizations, leading to financial losses, data breaches, and compromised security. This research study critically analyses features of email phishing through deep learning models. It examines the characteristics and patterns of phishing emails, such as misleading URLs, suspicious attachments, and linguistic cues that aim to manipulate the recipients. Then, the paper investigates the application of Artificial Neural Networks (ANNs) for analysing and detecting email phishing attacks. For experimentation, three ANNs, i.e. two recurrent, Long Short-Term Memory (LSTM) & Recurrent Neural Network (RNN), and one feed-forward, Convolutional Neural Network (CNN) are implemented and evaluated. The models are trained on the widely known Enron dataset of email samples, encompassing both spam and ham emails. The results show these ANNs outperform the traditional machine learning approaches as applied by contemporary research. While analyzing intra-model comparison using metrics, Recall & Type II Error, it is seen that CNN performs better than LSTM & RNN with 99% & 0.20% as compared to 31.32% & 38.60%, and, 52% & 27% respectively. Hence, the comparative analysis shows that the feed-forward (CNN) model performs better than recurrent (LSTM and RNN) models for detecting email phishing attacks. This finding can provide guidance to the research in this particular domain of email phishing.
KW - artificial neural networks
KW - convolutional neural network
KW - email phishing
KW - long short-term memory
KW - recurrent neural network
UR - http://www.scopus.com/inward/record.url?scp=85206094884&partnerID=8YFLogxK
U2 - 10.1109/Cyber-RCI59474.2023.10671515
DO - 10.1109/Cyber-RCI59474.2023.10671515
M3 - Conference contribution
AN - SCOPUS:85206094884
T3 - 2023 Cyber Research Conference - Ireland, Cyber-RCI 2023
BT - 2023 Cyber Research Conference - Ireland, Cyber-RCI 2023
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2023 Cyber Research Conference - Ireland, Cyber-RCI 2023
Y2 - 24 November 2023
ER -