基于扩张矩阵的模糊特征选择算法  

ALGORITHM OF FUZZY FEATURE SELECTION BASED ON EXTENSION MATRIX

作　　者：李云[1] 吴中福[1] 叶春晓[1] 李季[1] 刘嘉敏[2] 

机构地区：[1]重庆大学计算机学院,重庆400044 [2]重庆大学光电工程学院,重庆400044

出　　处：《模式识别与人工智能》2004年第4期417-423,共7页Pattern Recognition and Artificial Intelligence

基　　金：重庆大学骨干教师资助项目(No.2003A13)

摘　　要：在模糊分类器系统中,通常要将模式的原始特征投影到模糊空间,在模糊空间上进行特征选择,并在此基础上构建模糊决策规则.本文在对原始特征模糊化的基础上提出了一种构造模糊扩张矩阵的方法,它结合了常规扩张矩阵的优点和模糊集的特性,然后设计了一种基于模糊扩张矩阵的求解当特征取值为隶属度时的最优模糊特征子集的启发式算法,并从理论上证明了其正确性,也通过现实世界的数据集验证了它的较高效率.同时,该算法对基于集理论的模糊集相似性度量公式具有一定的鲁棒性.In the system of fuzzy classifier, we often adopt the language labels (such as high, low, narrow, wide, et al) to split the original feature into several fuzzy features, which are corresponding to fuzzy sets, and then select the optimal fuzzy feature subsets from the fuzzy space to construct the fuzzy decision rules. In the paper, firstly it describes the projection of features from original data onto fuzzy space through the selected fuzzy sets along with the transformation of numerical value of original features to the membership degree via the selected membership function. Subsequently, on the basis of the transformation, we design the fuzzy extension matrix, which integrates the superiorities of common extension matrix and the characters of fuzzy set. Finally, the paper presents a heuristic algorithm for the optimal fuzzy feature subset selection using fuzzy extension matrix and throws light on the steps of it by examples. The application of this algorithm is demonstrated by reducing the number of fuzzy features and the computational complexity used for the three real world datasets, which are well-known machine learning datasets. At the same time, it is proved to be insensitive to the similarity measures of fuzzy set based on the set theory.

关 键 词：模糊集 特征选择 扩张矩阵 

分 类 号：O159[理学—数学]