Abstract

Recent breakthroughs in deep learning algorithms have enabled researchers and practitioners to make significant progress in various hard computer science problems and applications from computer vision and perception, natural language processing and interpretation to complex reasoning tasks such as playing board games (e.g., Go, Chess, etc.) and even overthrowing human champions. Considering the expected acceleration and increase in computer threats, in this article, we explore the utility and capability of deep learning algorithms in the important area of network intrusion detection. We apply and compare various state-of-the-art deep learning frameworks (e.g., Keras, TensorFlow, Theano, fast.ai, and PyTorch) in detecting network intrusion traffic and also in classifying common network attack types using the recent CSE-CIC-IDS2018 dataset. Experimental results show that fast.ai, a highly opinionated wrapper for PyTorch, provided the highest accuracy of about 99% with low false positive and negative rates in both detecting and classifying various intrusion types. Our results provide evidence of the utility of various deep learning frameworks detecting network intrusion traffic.