Abstract: Feature selection is NP-Hard problem. In order to get a minimal feature subset of an information system, so-called Similar Extension Matrix(SEM) is defined to discriminate all objects by its elements, and then condensed to Condensed SEM(CSEM) by the included relation in logic. At Last, by means of the statistical values as heuristic information, a minimal feature subset is efficiently obtained in CSEM. The heuristic algorithm of minimal feature subset selection is proved very efficient by theoretical analysis and experiment.

Key words: Information system, Feature selection, Heuristic information, Extension matrix

摘要： 特征选取是一个NP-Hard问题。为了快速完成信息系统的一个最小特征选取，引入了类扩张矩阵的定义。通过类扩张矩阵的元素表示对象的差异，并利用逻辑上包含关系，有效浓缩类扩张矩阵。最后，以类扩张矩阵的统计信息为启发式信息，在浓缩类扩张矩阵中实现一个最小特征子集的快速求解。通过理论分析和实验，证明了该特征选取方法的高效性。

关键词: 信息系统, 特征选取, 启发式信息, 扩张矩阵

CLC Number: 

• TP311