不完备信息系统的容差邻域熵和属性选择  

Tolerance Neighborhood Entropy and Feature Selection in Incomplete Information Systems

作　　者：曾凯[1] 佘堃[1] 

机构地区：[1]电子科技大学计算机科学与工程学院,成都611731

出　　处：《小型微型计算机系统》2014年第5期1120-1123,共4页Journal of Chinese Computer Systems

基　　金：四川省科技厅项目(2012GZ0061)资助;国家自然科学基金项目(61300192)资助

摘　　要：邻域粗糙计算模型可以处理信息系统中名义属性和字符属性共存的问题而得到了广泛应用.现有的邻域粗糙计算方法仅讨论了其处理完备信息系统问题,然而实际应用中的数据往往是不完备的.针对这一问题,首先提出了可用于度量不完备信息系统的容差邻域熵;然后得出了一系列相关定义和性质,证明了容差邻域熵是香农熵在不完备信息系统上的自然推广;最后给出了基于容差邻域熵的属性选择算法.实验结果表明,所提出的算法避免了数据过分预处理而带来的冗余信息,使样本在算法选择的特征空间内保持了较高的分类精度,可以更好地处理信息系统不完备的问题.Neighborhood rough set theory has proven already to deal with nominal attributes and numerical attributes simultaneously. It is an efficient tool for knowledge discovering in complete information systems. However, some of types of the data are incomplete in practical environments. To solve this problem, first, we proposed the concept of the tolerance neighborhood entropy for uncertainty measures in incomplete information systems. Moreover, a number of definitions and properties are obtained. We generalize Shannon entropy to the tolerance neighborhood entropy in incomplete information systems. Finally, a heuristic algorithm based on tolerance neighborhood entropy is constructed to select features in incomplete information systems. The experimental results shows that this al- gorithm performed well both in feature selection and classification.

关 键 词：不完备信息系统 容差邻域熵 不确定性 属性选择 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]