Abstract: Recently，spectral clustering has wide application in data mining.Outlier detection detects and analyzes outliers in order to find information.The NJW of spectral clustering algorithm is applied in data mining combined with outlier factor successfully，and a new outlier detection algorithm based on spectral clustering is proposed.Experimental results demonstrate this algorithm has better efficiency and wide applicability compared with the original outlier detection algorithm which based on clustering.

Key words: NJW, outlier detection, spectral clustering

摘要： 最近几年，谱聚类思想开始用于数据挖掘领域，并取得了较好的效果；离群数据挖掘是对离群点进行检测，发掘出有用知识。将谱聚类中的NJW算法成功应用到离群数据挖掘领域，并结合离群指数的概念，提出了一种适合离群数据挖掘的谱聚类算法。与原有的基于聚类的离群检测算法相比，具有更好的效率和适应性。实验验证了所提算法的有效性和可行性。

关键词: NJW, 离群数据挖掘, 谱聚类