Malicious URL classification using artificial fish swarm optimization and deep learning

• Main Authors: Mustafa Hilal, Anwer, Hassan Abdalla Hashim, Aisha, G. Mohamed, Heba, K. Nour, Mohamed, M. Asiri, Mashael, M. Al-Sharafi, Ali, Othman, Mahmoud, Motwakel, Abdelwahed
• Format: Article
• Language: English; English
• Published: Tech Science Press 2023
• Subjects: TK7885 Computer engineering
• Online Access: http://irep.iium.edu.my/101886/7/101886_Malicious%20URL%20classification%20using%20artificial%20fish%20swarm.pdf; http://irep.iium.edu.my/101886/13/101886_Malicious%20URL%20classification%20using%20artificial%20fish%20swarm_SCOPUS.pdf; http://irep.iium.edu.my/101886/; http://doi.org/10.32604/cmc.2023.031371

Cybersecurity-related solutions have become familiar since it ensures security and privacy against cyberattacks in this digital era. Malicious Uniform Resource Locators (URLs) can be embedded in email or Twitter and used to lure vulnerable internet users to implement malicious data in their systems. This may result in compromised security of the systems, scams, and other such cyberattacks. These attacks hijack huge quantities of the available data, incurring heavy financial loss. At the same time, Machine Learning (ML) and Deep Learning (DL) models paved the way for designing models that can detect malicious URLs accurately and classify them. With this motivation, the current article develops an Artificial Fish Swarm Algorithm (AFSA) with Deep Learning Enabled Malicious URL Detection and Classification (AFSADL-MURLC) model. The presented AFSADL-MURLC model intends to differentiate the malicious URLs from genuine URLs. To attain this, AFSADL-MURLC model initially carries out data preprocessing and makes use of glove-based word embedding technique. In addition, the created vector model is then passed onto Gated Recurrent Unit (GRU) classification to recognize the malicious URLs. Finally, AFSA is applied to the proposed model to enhance the efficiency of GRU model. The proposed AFSADL-MURLC technique was experimentally validated using benchmark dataset sourced from Kaggle repository. The simulation results confirmed the supremacy of the proposed AFSADL-MURLC model over recent approaches under distinct measures