Transfer Learning Based Intrusion Detection Systems Using Recursive Feature Elimination

The rapid proliferation of new technologies such as cloud computing, Internet of Things (IoT), smart devices, and virtualization has resulted in overwhelming network traffic. Thus, companies must have strong defense measures to safeguard sensitive data from illegal invasion, which can lead to huge financial damage. Current intrusion detection systems (IDS) need additional improvements to decrease the occurrence of false positives and boost overall precision. Therefore, this paper proposes an integrated IDS by combining Recursive Feature Elimination to reduce the useful features. The influential features are then passed to the transfer learningbased Xception model to classify the images. We evaluated our proposed model on the UNSW-NB 15 dataset and obtained an overall accuracy of 97 %. Furthermore, the proposed strategy was analyzed and compared with existing IDS.