A Novel Approach to Multi-Classifier Based on Multiple Feature Sets with SVM for Network Intrusion Detection
| Autor: | Cheng-Yang Lai, 賴政揚 |
|---|---|
| Rok vydání: | 2007 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 95 When we detect and classify the same dataset or data with different classifier, inconsistent problem among detecting results may emerges. This shows that complementarity exists between these classifiers. In this thesis, we proposed a novel application of the data fusion technology to intrusion detection system. We tried to utilize its integration feature to show the complementarity among the classifiers so that the detection rate, classification accuracy could be increased and the false alarm rate could also be reduced with intrusion detection system. In this experiment, we used KDD cup' 99 as the dataset and Support Vector Machine (SVM) as the core classification tool. By cooperating five feature selection methods (Principal Component Analysis, PCA; Multiple Linear Regression, MLR; Discriminant Analysis, DA; Rough Set Theory, RST; Genetic Algorithms, GAs), five kinds of the examined feature sets were derived. Each feature set was trained by SVM so as to construct a classifier. Finally, data fusion was implemented by the Bayesian average method to integrate the classified results derived from the aforementioned classifiers so as to solve the inconsistent problem. According to our experiment, the proposed approach of data fusion to intrusion detection is proven to be feasible in the multi-classifiers framework. The accuracy is obviously higher than that of each single classifier. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|