Malicious URL Detection Model

Reyes Campos J.E.; Venegas Villarreal A.M.; Mendoza De Los Santos A.C.; González Vásquez J.A.; Escobedo Moreno E.D.R. · CISCI 2023 - Vigesima Segunda Conferencia Iberoamericana en Sistemas, Cibernetica e Informatica, Vigesimo Simposium Iberoamericano en Educacion, Cibernetica e Informatica, SIECI 2023 - Memorias (2023) · DOI: 10.54808/cisci2023.01.247

Resumen

This article presents an innovative approach to address the problem of malicious Uniform Resource Locator (URL) detection to protect users from online scams. The main goal of this research is to develop a model capable of identifying fraudulent URLs with high accuracy, classifying them into four distinct categories: benign, defacement, malware, and phishing. To develop the model, machine learning-based methods and a pre-trained BERT (Bidirectional Encoder Representations from Transformers) model for feature extraction were used. A large dataset containing labeled URL addresses, including examples from these four categories, was collected and processed. These data were used to train and fine-tune the model so that it could learn patterns and key features associated with each category. The developed model analyzes the structural features of URLs and textual content for classification with the help of BERT. Feature extraction techniques such as length analysis, presence of suspicious characters, domain and subdomain analysis, etc., were used. The result was a high-precision model with a 96.9% accuracy rate in detecting malicious web addresses. The model can effectively distinguish benign URLs from fraudulent ones and can also successfully classify malicious URLs into the subcategories of defacement, malware, and phishing. It is expected that the model will have a positive impact on cybersecurity.

Instituciones: Universidad Nacional de Trujillo; Universidad César Vallejo