Dugat-LSTM: Deep learning based network intrusion detection system using chaotic optimization strategy

Abstract

Network intrusion is a huge harmful activity to the privacy of the data sharing network. The activity will result in a cyber-attack, which causes damage to the system as well as the user’s data. Unauthorized activities such as data tampering, illegal access to data and theft of credentials are increasing on the internet world every day. The detection of intrusion may be done by multiple methodologies; still, it is the biggest issue in the networks. Hence, an automated attack classification model is required to promote classification accuracy with fewer error possibilities based on the input parameters. To get relief from the insecurity of data, this paper presents an innovative model using deep networks. The proposed model is a deep learning based network intrusion detection system using a chaotic optimization strategy. The method is pre-processed using data cleansing and M-squared normalization. After pre-processing, the unbalanced datasets are balanced using the Extended Synthetic Sampling approach. After balancing, the features of the dataset are taken out using kernel-assisted principal component analysis. The optimal features are selected by the Chaotic Honey Badger optimization algorithm. After all required features have been extracted, the attacks are classified by the Gated Attention Dual Long Short Term Memory (Dugat-LSTM). The above process is performed using the TON-IOT and NSL-KDD datasets. The prototype is evaluated using the following metrics: accuracy, precision, recall, and F1 score. The accuracy value of the proposed model is 98.76% in the TON-IOT dataset and 99.65% in the NSL-KDD dataset. Thus, the accuracy and robustness of the model show that it outperforms other existing models.