An Intrusion Detection System Based on Convolutional Neural Network for Imbalanced Network Traffic

Abstract

As the Internet integrates with social life closely, various cyber threats pose a huge challenge to Intrusion Detection Systems (IDS). The performance of IDS based on traditional machine learning did not meet our expectations. In this paper, we propose an intrusion detection model based on Convolutional Neural Network (CNN). Before CNN training, Synthetic Minority Oversampling Technique combined with Edited Nearest Neighbors (SMOTE-ENN) algorithm is applied to balance the network traffic. We use NSL-KDD dataset to evaluate the model. The proposed SMOTE-ENN-based CNN IDS model achieves an accuracy of 83.31%. Furthermore, the detection rates of User to Root (U2R) and Remote to Local (R2L) attacks are significantly improved. Results show that SMOTE-ENN-based CNN IDS outperforms the previous IDS model.