Improving Network Intrusion Detection Using Geometric Mean LDA

Elkhadir Zyad ； Chougdali Khalid ； Benattou Mohammed

《International Journal of Network Security》 20卷5期 (2018/09) Pp. 820-826

https://doi.org/10.6633/IJNS.201809_20(5).02

摘要

Anomaly based Intrusion Detection System (IDS) recognizes intrusion by adapting itself to identify normal behavior of the network. It then raises an alarm whenever any suspicious network behaviors are observed. Nonetheless, this kind of IDS is usually prone to small detection rate and high false positive rate due to difficulties involved in building normal network traffic pattern or a model. To avoid as much as possible this issue, many papers exploited a feature extraction method called linear discriminant analysis (LDA) as an intermediate step before constructing the model. Unfortunately, LDA has an important weakness, the class mean vector employed in this method is always estimated by the class sample average. That is not sufficient to provide an accurate estimate of the class mean, particularly with the presence of outliers. In this paper, to overcome that, we propose to use the geometric mean to estimate the class mean vector in LDA modeling. Many experiment on KDDcup99and NSL-KDD indicate that the proposed approach is more effective than numerous LDA algorithms.

關鍵字

Geometric Mean ； KDDcup99 ； LDA ； Network Anomaly Detection ； NSL-KDD