Network intrusion detection based on Contractive Sparse Stacked Denoising Autoencoder

The rapid growth of network scale leads to the increasingly prominent network security problems. Intrusion detection is an important method to resist complex and growing network attacks. For traditional shallow intrusion detection methods can not effectively identify and classify network intrusion d...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:IEEE International Symposium on Broadband Multimedia Systems and Broadcasting S. 1 - 6
Hauptverfasser: Lu, Jizhao, Meng, Huiping, Li, Wencui, Liu, Yue, Guo, Yihao, Yang, Yang
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: IEEE 04.08.2021
Schlagworte:
autoencoder
Broadcasting
characteristic dimension reduction
Data models
intrusion detection
Network intrusion detection
Network security
Noise reduction
Resists
Robustness
softmax
sparse denoising
ISSN:2155-5052
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The rapid growth of network scale leads to the increasingly prominent network security problems. Intrusion detection is an important method to resist complex and growing network attacks. For traditional shallow intrusion detection methods can not effectively identify and classify network intrusion data, this paper proposes a Contractive Sparse Stack Denoising Autoencoder(CSSDAE), which cascades multiple traditional Autoencoders, and introduces noise, sparse constraint and contractive penalty term on this basis, so as to improve the robustness of the model, enhance decoding ability of the deep network and promote intrusion detection performance. In addition, this paper improves the softmax classifier to make the feature vectors as compact as possible within the class and separate as much as possible between classes, and solves the problem of uneven number of sample classes by weighting. The experimental results show that compared with traditional AE, the CSSDAE's accuracy of network intrusion detection is effectively improved.
ISSN:2155-5052
DOI:10.1109/BMSB53066.2021.9547087