Intrusion detection technology based on CEGA-SVM

Author

Wei, Yuxin ; Wu, Muqing

Author_Institution

Institute of Communication Networks Integrated Technique BUPT, Beijing, China

fYear

2007

fDate

17-21 Sept. 2007

Firstpage

244

Lastpage

249

Abstract

In order to improve the classification accuracy and reduce the detection time, the optimization of feature extraction and SVM training model is combined together. In the procedure of feature extraction using CEGA with adaptive crossover and mutation, fitness of the individual is evaluated by the correct classification rate and conditional entropy. The optimization of SVM training model is processed at the same time with the feature extraction in order to find the best combination of optimal feature subset with the SVM training model. Results of the experiment using KDD CUP99 data sets demonstrate that applying CEGA-SVM can be an effective way for feature extraction and intrusion detection.

Keywords

Communication networks; Entropy; Feature extraction; Genetic algorithms; Genetic mutations; Intrusion detection; Machine learning; Neural networks; Support vector machine classification; Support vector machines; conditional entropy; genetic algorithm; intrusion detection; optimal feature subset; support vector machine;

fLanguage

English

Publisher

ieee

Conference_Titel

Security and Privacy in Communications Networks and the Workshops, 2007. SecureComm 2007. Third International Conference on

Conference_Location

Nice, France

Print_ISBN

978-1-4244-0974-7

Electronic_ISBN

978-1-4244-0975-4

Type

conf

DOI

10.1109/SECCOM.2007.4550339

Filename

4550339