A Novel Clustering Algorithm Based on One-Class SVM
Xinyu Huang, Xiaoyun Chen · 2009
One-class Support Vector Machine (OC-SVM),which is proposed to deal with the problems of classification ,intends to find the smallest hyper-sphere containing the positive data. As for the test point, one-class svm only judges it whether the test point belongs to that cluster. So OCSVM is often used in anomaly detection. But in the algorithm proposed in this paper, we first adopt shared nearest neighbor algorithm based on the kernel method (KSNN) to pre-cluster the input data, and then use weight of each point, which is produced by KSNN, to cluster through OCSVM.Experimental results show that our algorithm can deal with some irregular distributed data and high-dimension data effectively.