Attack on Network Traffic Classification

Abstract

Various network traffic management and intrusion detection solutions use network traffic classification. Machine Learning (ML), while deep learning (DL)-based models, had exhibited excellent performance in Internet traffic classification. Even though most services encrypt their communication, some modify their port numbers. Deep neural networks (DNNs) and other machine learning models are subject to adversarial attacks. Adversarial examples include applying a minor disturbance to the input data to force a taught classifier to misclassify the input while the human observer adequately identified it. The attacker and defense industries are interested in detecting disturbance since it has caused significant damage and has evolved into threats to computer and Internet users. Machine learning-based technique has been successfully deployed in perturbation detection in recent years. Other feature representations assist the training samples, and different classifiers are created to support them. As Adversarial machine learning (AML) is still under study, researchers have not attempted to train the model on the header part of the network traffic for classification.