Detecção de ataques DDoS e DoS em redes utilizando arquitetura Transformers de aprendizado de máquina

Abstract

Denial of Service (DoS) and Distributed Denial of Service (DDoS) attacks have been a major issue on the internet for a long time, affecting a broad segment of society. These attacks can disrupt network traffic in an online video game match and even cause millions in losses for large companies. To combat increasingly sophisticated and large-scale attacks, various solutions are being explored, including Artificial Intelligence (AI). The data was extracted from the CIC-IDS2017 dataset, properly structured and filtered to remove overly dispersed addresses as well as null and labeled addresses. A value of 1 was assigned to benign addresses and 0 to malicious ones. The data was then split into training and validation sets, tokenized, and fed into a Generative Pre-trained Transformer (GPT) based transformers architecture for batch training. Subsequently, validation was performed using 20% of the data, achieving an accuracy of 99.3%, demonstrating that the model is effective in detecting malicious addresses.