Detection of Malicious Network Traffic using Convolutional Neural Networks

Networks are typically exposed to various attacks such as Denial of Service, Shellcode, and Fuzzers with increasing connectivity of users and organizations. It is crucial to detect such malicious network traffic to alleviate their impact on organizations and provide security administrators with automatic alerts. While conventional machine learning classifiers can solve the binary classification problem of network traffic being normal or anomalous, they do not perform well when detecting multiple attack categories. This work addresses the issue of malicious network traffic detection using deep convolutional neural network architectures on the modern complex and challenging UNSW-NB15 dataset. The proposed work shows a significant improvement relative to existing techniques even for the difficult attack categories of Analysis, Backdoor, Shellcode, and Worms.