Application of Support Vector Clustering Algorithm to Network Intrusion Detection

The support vector clustering (SVC) algorithm is inspired from support vector machines (SVM). It takes the form of quadratic programming and can yield a global optimum, and gives a sparse representation of the data set by way of only a few number of support vectors. Aiming at its defection of slowly training of large-scale sets, and considering the characteristic of network intrusion detection, an improved algorithm of SVC is proposed in this paper. The proposed method uses similarity measurement instead of Euclidian distance, and makes small clusters substituted by their reference points, then compensates the information distortion caused by the references. The training of the SVC is enhanced on large-scale data set as well as unevenly distributed data set