Research Article
Using Long-Short-Term Memory Based Convolutional Neural Networks for Network Intrusion Detection
@INPROCEEDINGS{10.1007/978-3-030-06158-6_9, author={Chia-Ming Hsu and He-Yen Hsieh and Setya Prakosa and Muhammad Azhari and Jenq-Shiou Leu}, title={Using Long-Short-Term Memory Based Convolutional Neural Networks for Network Intrusion Detection}, proceedings={Wireless Internet. 11th EAI International Conference, WiCON 2018, Taipei, Taiwan, October 15-16, 2018, Proceedings}, proceedings_a={WICON}, year={2019}, month={1}, keywords={Intrusion detection system Deep learning Long-short term memory NSL-KDD dataset}, doi={10.1007/978-3-030-06158-6_9} }- Chia-Ming Hsu
He-Yen Hsieh
Setya Prakosa
Muhammad Azhari
Jenq-Shiou Leu
Year: 2019
Using Long-Short-Term Memory Based Convolutional Neural Networks for Network Intrusion Detection
WICON
Springer
DOI: 10.1007/978-3-030-06158-6_9
Abstract
The quantity of internet use has grown dramatically in the last decade. Internet is almost available in every human activity. However, there are some critical obstacles behind this massive development. Security becomes the hottest issue among the researchers. In this study, we focus on intrusion detection system (IDS) which is one of the solutions for security problems on network administration. Since intrusion detection system is a kind of classifier machine, it is allowed to engage with machine learning schemes. Related to this reason, the number of studies related to utilizing machine learning schemes for intrusion detection system has been increased recently. In this study, we use NSL-KDD dataset as the benchmark. Even though machine learning schemes perform well on intrusion detection, the obtained result on NSL-KDD dataset is not satisfied enough. On the other hand, deep learning offers the solution to overcome this issue. We propose two deep learning models which are long-short-term memory only (LSTM-only) and the combination of convolutional neural networks and LSTM (CNN-LSTM) for intrusion detection system. Both proposed methods achieve better accuracy than that of the existing method which uses recurrent neural networks (RNN).