Abstract
This work proposes an end-to-end model for network attack detection and network attack classification using deep learning-based recurrent models. The proposed model extracts the features of hidden layers of recurrent models and further employs a kernel-based principal component analysis (KPCA) feature selection approach to identify optimal features. Finally, the optimal features of recurrent models are fused together and classification is done using an ensemble meta-classifier. Experimental analysis and results of the proposed method on more than one benchmark network intrusion dataset show that the proposed method performed better than the existing methods and other most commonly used machine learning and deep learning models. In particular, the proposed method showed maximum accuracy 99% in network attacks detection and 97% network attacks classification using the SDN-IoT dataset. Similar performances were obtained by the proposed model on other network intrusion datasets such as KDD-Cup-1999, UNSW-NB15, WSN-DS, and CICIDS-2017.
[Display omitted]
•Recurrent model’s hidden layers feature fusion for network intrusion detection.•Ensemble meta-classifiers for network intrusion detection and classification.•Evaluation of models are shown on more than one network intrusion dataset.•Feature visualization using t-distributed stochastic neighbor embedding (t-SNE).