A GAN and Feature Selection-Based Oversampling Technique for Intrusion Detection

In recent years, there have been numerous cyber security issues that caused considerable damage to the society. The development of efficient and reliable Intrusion Detection Systems (IDSs) is an effective countermeasure against growing threats. modern high-bandwidth, large-scale network environments, traditional IDSs suffer from a high rate missed false alarms. Researchers introduced machine learning techniques into intrusion detection with good results. However, due scarcity attack data, such methods’ training sets are usually unbalanced, affecting analysis performance. this paper, we survey analyze design principles shortcomings existing oversampling methods. Based on findings, take perspective imbalance dimensionality datasets in field propose technique based Generative Adversarial Networks (GAN) feature selection. Specifically, model complex high-dimensional distribution attacks Gradient Penalty Wasserstein GAN (WGAN-GP) generate additional samples. We then select subset features representing entire dataset variance, ultimately generating rebalanced low-dimensional for training. To evaluate effectiveness our proposal, conducted experiments NSL-KDD, UNSW-NB15, CICIDS-2017 datasets. experimental results show method can effectively improve performance models outperform baselines.