Abstract: Granular computing theory is a new efficient method to deal with uncertain, incomplete and inconsistent knowledge. Knowledge granulation is one of important tools to deal with uncertain information in granular computing theory. Many existing algorithms of outlier mining mainly aim for certain data, very little work has been done for uncertain data aiming to outlier mining based on knowledge granulation. Therefore, after introducing knowledge granulation concept, relative knowledge granulation and outlier degree are defined for measuring the outlier data. A new algorithm for outlier mining based on knowledge granulation is proposed. This algorithm can effectively obtain outliers from data set. The validity of the algorithm is depicted by an example.

Key words: granular computing, rough sets, outlier data mining, knowledge granulation

摘要： 粒计算理论提供了一种新的处理不确定、不完全与不一致知识的有效方法。知识粒度是粒计算理论中度量不确定信息的重要工具之一。已有的异常数据挖掘算法主要针对确定性的异常数据挖掘，采用知识粒度度量不确定性数据，进行异常数据挖掘的研究尚未报道。为此，在引入知识粒度概念的基础上，定义了相对知识粒度及异常度来度量数据之间的异常程度，并提出基于知识粒度的异常数据挖掘算法，该算法可有效进行异常数据的挖掘。实例验证了该算法的有效性。

关键词: 粒计算, 粗糙集, 异常数据挖掘, 知识粒度