基于KNN图的两阶段孤立点检测及应用研究

Novel two-stage approach based on KNN graph for outlier detection and its application research

针对两种基于KNN图孤立点检测方法:入度统计法(ODIN)和K最邻近(K-nearest Neighbor,RSS)算法的不足,提出了一种新的改进方法:两阶段孤立点检测方法,并进行了适当扩充使之适用于数据集中孤立点数目未知情况下的孤立点检测。算法应用于"小样本,高维度"的基因微阵列数据集进行样本孤立点检测取得了很好效果,证明了此方法的有效性。

Aiming at overcoming the shortcoming of two KNN graph based outlier detection methods：Outlier Detection using Indegree Number （ODIN） algorithm and K-nearest neighbor （RSS） algorithm,this paper proposes a novel improved approach：twostage KNN graph based outlier detection method.This method can be employed to detect the outliers of datasets with the number of outliers being unknown.Appling it into the ＂small sample,high dimension＂ microarray datasets achieves a good result.