基于序关系的非完备信息案例推理集成方法研究

RESEARCH ON ENSEMBLE METHOD OF CASE-BASED REASONING WITH INCOMPLETE INFORMATION BASED ON ORDERING RELATION

针对案例推理系统中数据集存在数据缺失的非完备信息问题,利用序关系基本原理,设计了案例推理集成方法(ORCBR)。通过对非完备信息下确定符号属性、确定数值属性、区间数值属性以及模糊语言属性等属性间相似性度量的研究,计算出目标案例与历史案例的相似性矩阵。在此基础上,利用序关系构建了相似性矩阵中不同属性的集成排序算法,从而得到最相似历史案例。通过对UCI数据库中非完备信息数据集的测试表明,OR-CBR方法比经典案例推理方法准确率高、效率高,很好地解决了非完备信息数据集的案例推理问题。

A case-based reasoning ensemble method （OR-CBR） is put forward to solve the problem of data missing in the case-based reasoning system, utilizing order relation theory. The similarity measurements of incomplete information attributes, including crisp symbol, crisp data, interval number and fuzzy linguistic, has been respectively studied. Then, the similarity matrix between target case and historical case is ealculated. On the basis of this, an integrate sorting algorithm is designed for each attribute in similarity matrix in order to get the most similar historical case. The test results on incomplete data sets of the UCI show that the accuracy rate and efficiency of OR-CBR method are obviously higher thanother classical algorithms, solving the problem of incomplete information data sets in case-based reasoning system.