Using KPCA feature selection and fusion for intrusion detection
Rui-xia Zhang, Guojian Zhi · 2010 Sixth International Conference on Natural Computation · 2010
The main task of intrusion detection is to extract meaningful and effective features from redundant and noisy features causing poor detection accuracy. A method of feature selection by Kernel Principle Component Analysis (KPCA) and fusion is proposed. The basic features, contend features and traffic features are extracted respectively by KPCA. Then we use two levels of fusion (feature fusion and decision fusion) technique to improve the performance of intrusion detection system. However, simple combining features will not work as well as expected. For this reason, a new feature fusion method, IS feature fusion is presented. Experiments have been done on dataset in KDD-99 and simulation results show that our method by KPCA feature selection is an effective and IS feature fusion outperforms other fusion techniques.