基于属性约简和相对熵的离群点检测算法

本研究结合信息熵与粗糙集理论中的属性约简技术,提出了一种新颖的离群点检测算法。这种方法通过在更小的属性子空间去获得相同或相近的离群数据集,使对离群数据的分析更加集中于较小的目标域。该算法对原属性空间进行划分,通过分析计算将具有最大相对熵与负相对势的对象集合判定为离群点集合。为了验证算法的有效性,还在通用数据集上进行了测试,理论分析和实验结果表明该离群点检测算法是有效可行的。

An outlier detection algorithm based on attribute reduction and relative entropy

A new outlier detection algorithm combining a rough set and information entropy technology was proposed. This approach could obtain similar outlier sets by means of searching in an attributes subspace, which could lead the analysis of outlier detection to focus better on narrow and specific object fields. This algorithm divided the original attribute space into several segments, which filtered out those subjects with largest relative entropy negative relative cardinality as the outliers. To prove this algorithm’s effectiveness, experiments on a real world dataset were conducted. Theoretical analysis and experimental results showed that this method of outlier detection was efficient and effective.