Intrusion Detection Using Temporal Convolutional Networks

Intrusion detection system is an important network security facility. With the fast development of information technology, the information security is getting more serious. On the other side, making the IT equipment more intelligent via AI methods becomes a research hotpot. Recent studies show that temporal convolutional networks can outperform recurrent networks and convolutional architectures in sequence modeling problems. In this work, we propose a data processing method for intrusion detection. We conduct a systematic evaluation of temporal convolutional networks for intrusion detection with NSL KDD data set. Compared with other standard baseline machine learning methods and some advanced deep learning architectures, the proposed model gives a promising performance in different level tests. With limited computational cost, TCN model converges fast and shows good performance. The proposed model can be easily adjusted to raw inputs and can be extended to large-scale online applications.