ISAD:一种新的基于属性距离和的孤立点检测算法

An Algorithm for Outlier Detection Based on the Sum of Attribute Distances

孤立点是数据对象在某些属性(维)上波动形成的。由此,本文提出了关键属性的概念,用于描述影响数据稳定性的属性。在真实数据集中,只有一部分属性是能够决定某数据是否是孤立点的关键属性。由此,本文提出了关键属性隶属度的定义及其求解算法,并在此基础上提出了一种新的基于属性距离和的孤立点检测算法。实验结果表明,该算法较基于单元的算法在效率及维数可扩展方面均有显著提高。

Outliers are the result that the data objects fluctuate in certain attributes（dimensions）. Therefore, we present the concept of key attribute to describe the attributes that affect data stability. In real datasets, only a few attributes are key attributes that can determine whether the data is an outlier or not. Thus we present the definition of the degree of membership for the key attributes and its correlative algorithm. Moreover, a new algorithm based on the sum of attribute distances is designed for outlier detection. The experimental results show that the new algorithm is effective, and its efficiency increases obviously.