Study of Deep learning Architectures for Network Intrusion Detection

The task of Network Intrusion detection involves detecting and stopping bad actors from obtaining unauthorized access to systems. Due to the varied nature of the network attack types, Multi-class classification models are routinely employed in Autonomous Network based Intrusion Detection systems (NIDS), for distinguishing malicious traffic from benign traffic. The performance of these Machine learning models on the task are often evaluated using standard datasets like UNSW-NB15 and TON_IoT. Gradient Boosted Decision Tree (GBDT) models have been routinely used by practitioners in Intrusion detection systems due to their superior performance on tabular data classification tasks. In this study, we compare the performance of three select deep learning models against GBDT models on NIDS. Specifically, we study the performance of Embedding augmented Multilayer Perceptron (MLP) model and Tabular data adaptations of Image classification model architectures on NIDS.