RANet: Network intrusion detection with group-gating convolutional neural network

Abstract

With the rapid increase of human activities in cyberspace, various network intrusions are tended to be frequent and hidden. Network intrusion detection (NID) has attracted more and more attention from industrial and academic fields. Over the years, researchers have developed artificial intelligence methods to tackle them. However, most existing methods are usually not feasible and sustainable when faced with the demands of current NID systems. To alleviate this problem, this paper proposes a novel convolutional neural network (CNN) named RANet for NID automatically. In RANet, we not only introduce a Group-Gating module but also apply the overlapping method to the last max-pooling layer. Based on the hyper-parameter settings of our RANet, a lot of performance comparison experiments are conducted. The results demonstrate that RANet achieves better NID performance than strong baselines and existing state-of-the-art methods on five publicly available NID benchmarks. For example, the RANet improves accuracy with approximately 5.3% on NSL-KDD Test−21 dataset through comparisons to state-of-the-art baselines. Moreover, the results of RANet also indicate that it has the great potential or use in current NID systems.