SPE-ACGAN: A Resampling Approach for Class Imbalance Problem in Network Intrusion Detection Systems

Network Intrusion Detection Systems (NIDSs) play a vital role in detecting and stopping network attacks. However, the prevalent imbalance of training samples traffic interferes with NIDS detection performance. This paper proposes resampling method based on Self-Paced Ensemble Auxiliary Classifier Generative Adversarial Networks (SPE-ACGAN) to address problem sample classes. To deal class problem, SPE-ACGAN oversamples minority by ACGAN undersamples majority SPE. In addition, we merged CICIDS-2017 dataset CICIDS-2018 into more imbalanced named CICIDS-17-18 validated effectiveness proposed using three datasets mentioned above. is effective than other methods improving particular, improved F1-score Random Forest, CNN, GoogLeNet, CNN + WDLSTM 5.59%, 3.75%, 3.60%, 3.56% after resampling.