Q266 : Automatic URL Generation using Deep Learning Techniques
Thesis > Central Library of Shahrood University > Computer Engineering > MSc > 2024
Authors:
Amirreza Moradi [Author], Mohsen Rezvani[Supervisor], Mansoor Fateh[Advisor]
Abstarct: Phishing attacks, one of the greatest cybersecurity threats in today's world, are conducted with the aim of deceiving users into disclosing personal information or committing financial fraud. Today, detecting and blocking phishing URLs is of utmost importance due to the substantial financial and security damages they can cause. One common method for carrying out these attacks is the use of fake URLs that seemingly lead to legitimate websites such as banks or online stores. Traditional phishing detection methods are baxsed on manual techniques that require significant human resources and constant pattern updates. In contrast, artificial intelligence-baxsed methods can automatically generate phishing URLs without human intervention. In this research, we explore methods for automatic generation of phishing URLs using deep learning techniques, particularly Generative Adversarial Networks (GANs). GANs consist of two neural networks that compete with each other: a generator that produces fake data such as phishing URLs, and a discriminator tasked with distinguishing real data from fake. In the proposed GAN architecture, the generator uses a GRU network, while the discriminator employs a CNN network. This method offers several advantages over traditional phishing URL generation methods. Firstly, it can generate phishing URLs automatically and at high speed. Secondly, it can produce URLs that closely resemble real URLs, making them difficult for humans to detect. However, the main objective of this research is to use this method to generate a blacklist of candidate phishing URLs for a given domain. This blacklist can serve as a powerful tool to protect users against phishing attacks. Finally, the presented GAN network is tested on our data, which consists of URLs extracted from three bank domains. The results indicate that this algorithm is capable of generating a powerful blacklist with an average domain similarity of over 68% for all domains, producing 72 URLs for Mellat Bank, 52 URLs for Tejarat Bank, and 30 URLs for Sepah Bank.
Keywords:
#Phishing Attacks #Phishing URL #Deep Learning #GAN networks #Blacklist Keeping place: Central Library of Shahrood University
Visitor: