A hybrid clustering algorithm based on ART2 and its application in anomaly detection

Adaptive Resonance Theory (ART) and k-means have been widely used for clustering, but those two algorithms have their own limitations. In this paper a hybrid clustering algorithm is proposed which is based on ART2 and k-means. Firstly ATR2 is executed to find the initial cluster numbers and initial cluster centers, k-means uses these values to initialize its parameters and find new cluster centers, then these new cluster centers are sent back to ART2, ART2 use them to initialize connection weights between F1 layer and F2 layer, and get the final improved clusters. To prove its effectiveness it was applied in intrusion detection. The KDDpsila99 data sets are used as experimental data. Experiments show that clustering results are improved.