Research Article
Towards an Attention-Based Accurate Intrusion Detection Approach
@INPROCEEDINGS{10.1007/978-3-030-91424-0_16, author={Arunavo Dey and Md. Shohrab Hossain and Md. Nazmul Hoq and Suryadipta Majumdar}, title={Towards an Attention-Based Accurate Intrusion Detection Approach}, proceedings={Quality, Reliability, Security and Robustness in Heterogeneous Systems. 17th EAI International Conference, QShine 2021, Virtual Event, November 29--30, 2021, Proceedings}, proceedings_a={QSHINE}, year={2021}, month={11}, keywords={Network security Intrusion detection Attention Neural network}, doi={10.1007/978-3-030-91424-0_16} }- Arunavo Dey
Md. Shohrab Hossain
Md. Nazmul Hoq
Suryadipta Majumdar
Year: 2021
Towards an Attention-Based Accurate Intrusion Detection Approach
QSHINE
Springer
DOI: 10.1007/978-3-030-91424-0_16
Abstract
With the advancement of 5G and IoT, the volume of network traffic is growing in a tremendous rate (e.g., 235.7 Exabytes (EB) in Internet traffic, a 3.2-fold increase from 2016), leading to an alarming rise in different types of attacks. As a result, the requirements of an intrusion detection system (IDS) are also evolving. In addition to having a large number of flow-based intrusion detection systems powered by machine learning techniques, achieving higher accuracy including higher recall and precision has become equally important. While most of the existing works successfully achieve accuracy, they still strive to achieve a good recall score or minimize the False Negative Rate (FNR) as well as the False Positive Rate (FPR). In this paper, we investigate the potential of combining the state-of-the-art neural network models (i.e., CNN, LSTM, and GRU) with attention mechanisms (where attention helps the model to selectively concentrate on more relevant factors) for improving the accuracy of intrusion detection systems. We evaluate our model with the most recent and state-of-the-art benchmark datasets (e.g., CSE-CIC-IDS-2018, and NSL-KDD) and compare the obtained results with the existing works. Empirical results show that our proposed model outperforms the existing works in terms of accuracy while achieving a higher recall score (e.g., a maximum recall of 100%, 99.91% for CSE-CIC-IDS-2018, and NSL-KDD datasets, respectively) and higher F1-Score (e.g., a maximum F-1 score of 100%, 99.22% for CSE-CIC-IDS-2018, and NSL-KDD datasets, respectively).
