Unsupervised Network Intrusion Detection Using Convolutional Neural Networks

Abstract

Unsupervised Machine Learning (ML) is more desirable than supervised ML-based network intrusion detection techniques. Convolutional Neural Network (CNN) performs excellently in tasks related to image processing and computer vision applications as a supervised learning (SL) model, but SL is not suitable for a zero-day attack detection for network intrusion detection (IDS) system. In this work, the power of CNN in conjunction with autoencoder (AE) is used to develop unsupervised machine learning techniques to detect anomalies in network traffic. Two models are developed: CNN-based pseudo-AE and CNN-based classical AE models. The PVAMU-DDoS2020 dataset is used for training and testing the models. The results show the models are efficient in detecting anomaly (distributed denial-of-service) traffic for the unseen traffic flows from the PVAMU-DDoS2020 in an unsupervised fashion.