Intrusion Detection System Based on Transformer

Abstract

The digital era has profoundly transformed the daily operations of both public and private sectors. They have embraced digital platforms across multiple devices, presenting diverse content to meet the challenges of everyday work. The shift to digital platforms has enabled paperless operations but has also brought new challenges. The rapid development of digital technology has given rise to malicious activities such as hacking, malware, phishing, and DDoS attacks. These threats aim to gather intelligence, steal sensitive data, or demand ransom, leading to a surge in cybersecurity incidents.
In this paper, we propose a monitoring classification model using a pre-trained Transformer-based BERT multi-classification model. This model can perform real-time online analysis of network traffic and identify potential anomalies or suspicious behavior. Proactive prevention and mitigation of malicious attacks involve understanding attack intent through network analysis. This study trains machine learning models on the CIC-IDS-2017 dataset, which includes various attacks and normal traffic. The proposed solution is based on a flow-based intrusion detection system that leverages machine deep learning with Transformers. The goal is to establish a real-time intrusion detection system capable of identifying threats through traffic analysis, providing a robust defense for the cybersecurity challenges of the digital era.