Abstract
Traditional network defense technologies such as firewalls are unable to detect the evolving types of attacks on networks, leading to the need for network intrusion detection systems (NIDSs) that provide better solutions. In this paper, we propose an effective deep learning method to NIDS based on a two-stage approach with a sparse autoencoder and a number of different classifiers, to create three models. The proposed approach uses the autoencoder for feature learning and dimensionality reduction, thereby reducing training and testing times. The new feature vector is then input into three classifiers, improving their detection capability for intrusion and classification accuracy. We study and compare our models with a number of other works in the literature as well as some state-of-the-art methods. Results show that our approach performs better than all other approaches in terms of detection rate, and comparably in terms of accuracy.