Malicious URL Detection Using Transformers’ NLP Models and Machine Learning

Nowadays, the WEB has become an indispensable part of our life. Indeed, the profound digital transformation initiated by all actors and accelerated by the health crisis has known the rise of multiple threats. Using malicious URLs, cybercriminals deceive their victims: individuals and organizations. By stealing their confidential information or tricking them into running malicious code, they can lead to losses of billions each year. Although various methods are available to detect these URLs, cybercriminals have developed new strategies to avoid detection. With the development of artificial intelligence, new promising automatic learning methods have emerged. In this paper, we propose a malicious URL detection method based on pre-trained transformers’ NLP models to extract features and we evaluate the effectiveness of various machine learning models. By combining the transformers’ NLP features with traditional lexical features, the best model can achieve a remarkable accuracy of 99.22%. Moreover, it can maintain a high level of robustness with 99.29% and 98,71%, respectively, for precision and recall scores.