Towards fast detecting intrusions: using key attributes of network traffic

Wei Wang (1) , Sylvain Gombault (2) , Thomas Guyet (1)

1 DREAM - Diagnosing, Recommending Actions and Modelling
2 RSM - Département Réseaux, Sécurité et Multimédia

Extracting attributes from network traffic is the first step of network intrusion detection. However, the question of which or what attributes are most effective for the detection still remains. In this paper, we employed information gain, wrapper with Bayesian Networks (BN) and Decision trees (C4.5) respectively to select key subsets of attributes for network intrusion detection based on KDD Cup 1999 data. We then used the selected 10 attributes to detect DDoS attacks in the real environments. The empirical results based on DDoS attack data collected in the real world as well as KDD Cup 1999 data show that only using the 10 attributes, the detection accuracy almost remains the same or even becomes better compared with using all the 41 attributes with both BN and C4.5 classifiers. Using a small subset of attributes also improves the efficiency in terms of attribute forming, models training as well as intrusion detection.

Mots clés: Intrusion detection; DDoS attack detection; Bayesian networks; C4.5; Attribute selection; Information gain

Domaines: Cryptographie et sécurité [cs.CR]; Réseaux et télécommunications [cs.NI]

Wei Wang, Sylvain Gombault, Thomas Guyet. Towards fast detecting intrusions: using key attributes of network traffic. ICIMP 2008 - Third International Conference on Internet Monitoring and Protection, Jun 2008, Bucharest, Romania. pp.86-91, ⟨10.1109/ICIMP.2008.13⟩. ⟨hal-01833587⟩