基于粗集理论的特征子集选择算法

1.引言长期以来,特征子集选择技术一直是机器学习领域中的关键难题之一。由于学习对象的多样性,尤其是新的系统随着应用的发展而不断涌现,使人们无法用某种特定的工具或方法来完全解决这一问题,新的特征子集选择技术仍然受到人们广泛关注。20世纪80年代初,波兰数学家Z.Pawlak提出一种新的理论工具——“粗集”,用于解决不完整和不精确信息的知识表达、学习及归纳等问题。这一理论的特点是:除了问题所需处理的数据之外,不需要额外提供任何外界信息或

Algorithms Based on Rough Set Theory for Feature Subset Selection

Rough set theory is a valid mathematical tool for dealing with the problem of feature subset selection. In this paper, to break the restriction of the conception of conditional entropy and provide an effective measurement to the relative importance of redundant features, system entropy of a decision system is defined based on rough set theory; some of its algebraic characteristics are also researched. Then two similar heuristic algorithms are introduced to select features based on the notion of system entropy. Moreover, different characteristics of the two proposed algorithms are also deeply analyzed and discussed. The two new algorithms may surely maintain the discernible relation of decision systems; their space and time complexities are obviously much lower than that of analogous algorithms in literature. Simulation results on numerous UCI machine-learning databases indicate that the optimal feature subsets may always be expected through the two algorithms on almost all cases.