Network Intrusion Detection Using Sequence Models

Abstract

The increase in network users has diversified the nature of attacks and increased their frequency. Existing intrusion detection systems rely on inefficient signature based approaches which can easily be evaded by attackers. Many shallow learning approaches have been explored but they require expert knowledge and longer training times. In this paper we utilize architectures such as RNN, LSTM and GRU to provide a solution to this problem. We also analyze and build upon an existing NDAE model and provide a comparative analysis. We have implemented our models using Keras with a TensorFlow backend. The benchmark NSL-KDD dataset is used for training and validation. The results obtained are promising and our models have potential to detect attacks in real-time backbone network traffic.