Abstract: This paper proposes a neighborhood density based outlier detection method, which is applicable to the data with mixed numerical and categorical features. In this method, the outlierness indicator of a sample is defined as the weighted sum of neighborhood cardinality and  average neighborhood density of this sample. To validate the effectiveness of this algorithm, it performs a series of experiments. Experimental results show that the proposal is applicable and effective to mixed data.

Key words: outlier detection, neighborhood density, mixed data

摘要： 提出了一个基于邻域密度的异常检测方法，它能处理混合数据的异常值。在该方法中，样本的异常指标被定义为该样本的邻域大小和该样本的平均邻域密度的加权和。为了验证提出的方法，进行了一系列实验。实验结果表明新提出的方法适用于混合数据，并且比其他检测方法更有效。

关键词: 异常检测, 邻域密度, 混合数据