作者: Tian Shengfeng , Wei Xiaotao , Huang Houkuan
DOI:
关键词:
摘要: A semi-supervised clustering algorithm based on the traditional k-means is proposed for network anomaly detection. We improve original mainly in three aspects. First, number of clusters automatically decided by merging and splitting clusters. Second, a small portion labeled samples are employed to supervise process stage. Also, we modify directly symbolic attribute values. Experimental result KDD 99 intrusion detection datasets shows that our has high rate while maintaining low false positive rate.