Abstract:

In traditional SVM based intrusion detection approaches, both core function construction and feature selection use prior knowdege. Due to this, they are not only inefficient but also inaccurate. It is observed that integrating information entropy theory into SVMbased intrusion detection can enhance both the precision and the speed. Concludely speaking, SVMbased entropy intrusion detection algorithms are made up of two aspects: on one hand, setting sample confidence vector as core function's constructor of SVM algorithm can guarantee the mapping relationship between training sample and optimization classification plane. Also, the intrusion detection’s maximum interval can be acquired. On the other hand, simplifying feature subset with samples's entropy as metric standard can not only shrink the computing scale but also improve the speed. Experiments prove that the SVM based entropy intrusion detection algoritm outperfomrs other tradional algorithms.

Key words: intrusion detection;SVM;entropy