KFDA-waveletcluster based intrusion detection technology

In this paper, we propose a new intrusion detection technology which combines feature extraction with wavelet clustering method. Our intrusion detection model setup has two phases, where the first phase is to project the input data into high dimensional space by using the discriminant vectors extracted by Kernel Fisher Discriminant Analysis. By using KFDA, we can reduce the dimension of the input data and make the dataset more separable. Then the second phase is to set up the detection model based on wavelet clustering. We extend original wavelet clustering algorithm for intrusion detection. Second transformation of feature space is processed by using wavelet transform for removing the outliers and making the boundary between clusters clear. Clusters are set up on the second transformed feature space and we label the clusters by using the similarity information between training datasets and clusters. Experiments using the KDD CUP99 dataset demonstrate that by combining KFDA and wavelet clustering can be an effective way for intrusion detection.