Ensemble Model Based on an Improved Convolutional Neural Network with a Domain-agnostic Data Augmentation Technique
Faraz Fatahnaie, Seyyed Amir Asghari, Armin Azhdehnia, Mohammadreza Binesh Marvasti · 2022
With the increase of online activities and the growing number of online services, various cyber threats pose a significant challenge to Network Intrusion Detection systems (NIDS). To face these threats, available imbalance sources made researchers develop resampling techniques to have a balance training process. In this paper, a domain-agnostic data augmentation approach followed by random under sampling is used to achieve credible generalized and robust IDS. Moreover, the framework benefits from the potentiality of deep learning models to extract more meaningful features. The final model of the paper was obtained after the ensemble of three improved convolutional neural networks. Each model is trained on a specific subset of the NSL-KDD dataset which is generated by the resampling method. The simulation results illustrate that the model achieves an accuracy of 83.3% which is 6.5% higher, when the original dataset is used. The codes are available at https://github.com/armin-azh/ensemble.