知识粒度框架下不确定性度量研究

Research on Uncertainty Measures Using Knowledge Granulation

不确定性度量是人工智能领域研究热点之一,它可以度量属性集的区分能力,并为属性约简提供了有效的度量工具。目前,已经提出了适用于信息系统的信息量和互补熵,应用于相容决策表的差别矩阵和类别特征矩阵,应用于任何决策表的可辨识的对象对、不可辨识的对象对以及互补条件熵等不确定性度量方法。然而,不同的不确定性度量对于并行属性约简算法效率并不相同。为此,从类内度量和类间度量两个角度研究了这些不确定性度量在知识粒度框架下差异和联系,通过实例验证了结论的正确性,这将为并行属性约简算法中不确定度量的选择提供了理论依据。

Uncertainty measure is one of the hot topics in the area of artificial intelligence,which can measure the capabilities of the attribute set and can be employed during the attribute reduction process.The information quantity and complement entropy are employed for the information systems,the discernibility matrix and class feature matrix apply to the consistent decision tables,while the discernibility object pair,indiscernibility object pair and complement condition entropy are suitable for any decision tables.However,different measures of uncertainty largely affect the efficiency of the parallel attribute reduction algorithms.To address this issue,from the perspective of the inter-class and intra-class measures,this paper studied systemically the interrelationships of classical measures of uncertainty,and transformed these uncertain measures into the forms of knowledge granulation.The example was employed to explain the results.It can help to guide how to choose the most appropriate uncertainty measure for the parallel attribute reduction algorithms.