Centro de Documentação da PJ
Analítico de Periódico
CD 358
UPADM [Recurso eletrónico] : a novel URL phishing attack detection model based on machine learning and deep learning algorithms / Saman Iftikhar, Omar Ahmed Abdulkader, Bandar Ali Al-Rami Al-Ghamdi
International Journal of Cyber Criminology, Vol. 18, n. 1 (January-June 2024), p. 244-260
Ficheiro de 624 KB em formato PDF.
PIRATARIA INFORMÁTICA, FRAUDE INFORMÁTICA, REDE NEURONAL ARTIFICIAL, SEGURANÇA INFORMÁTICA
Motivation: phishing attacks are among the most common and destructive cyberattacks. Attackers often use deceptive URLs to lure victims into revealing sensitive information. Traditional methods of detection usually fail with the increasing complexity of these attacks. It should reduce financial and reputational losses related to phishing attacks; train users on the dangers involved with such threats and promote safety online, contributing to assurance among users in safely using the internet without fear of theft or exposure. Solution proposed: the paper proposes an effective URL phishing attack detection model, UPADM, which can detect phishing URLs from legitimate ones by using the efficient and effective algorithms of Machine Learning and Deep Learning. Extracted features are drawn from three major sources, including lexical features of URLs, domain-based features, and website behavioral characteristics. It has been trained with a large set of legitimate and phishing URLs to give the best discrimination among them. Methodology: the model follows a hybrid architecture where the machine learning algorithms will do an initial filtering based on lightweight features, and deep learning networks perform an in-depth analysis using advanced feature representations. This layered approach increases detection accuracy without sacrificing efficiency and thus is suitable for real-time applications in various environments. Findings and their impact: extensive experiments were done on a high volume, highly varied dataset of benign and malicious URLs for performance evaluation of the proposed model. The results have shown that our model is superior to the current state-of-the-art techniques with the highest accuracies of 99.7% and almost 97% for precision, recall, and F1-score. The results show that the coupled ML-DL methods hold great promise toward obtaining a robust, scalable solution to handle phishing attacks. Novelty: the paper examines in depth the various strengths and shortcomings of several popular machine learning and deep learning approaches regarding the identification of phishing attacks by Uniform Resource Locator addresses with a view to providing broad information for the development of more robust security approaches.
