基于样本和特征搜索空间不断缩小的模糊粗糙集特征选择

为了提高模糊粗糙集特征选择算法的计算效率,在每次迭代过程中通过不断缩减样本和特征的搜索范围,提出了一种新的模糊粗糙集特征选择算法.为了减少样本的搜索范围,利用样本对决策类下近似隶属度的单调性,构建样本的筛选机制,用以筛除当前所选特征子集已保持决策类下近似隶属度的样本;为了缩减特征的搜索范围,采用特征冗余性概念,构建特征搜索机制,用以移除已被确定为冗余的特征;通过融合样本筛选机制和特征搜索准则,设计模糊粗糙集特征选择的高效算法.数值实验表明,所提算法具有高效性和有效性.

Fuzzy rough set-based feature selection with the ever-shortening scope of samples and features

In order to improve the time efficiency of fuzzy rough set-based feature selection methods, a novel fuzzy rough set-based feature selection algorithm is developed by step-by-step shrinking the search scope of both samples and features at each iteration of locating a best feature. For the reduction of the sample search range, the sample filtering mechanism is first constructed by the monotonicity of the membership degree of a sample belonging to the lower approximation of its decision class. During the iteration of selecting a best feature, the sample filtering mechanism can ignore samples, of which the degree belonging to lower approximations of their decision classes is preserved by a current selected feature subset. These earlier samples will not be considered during the later iteration of selecting other best features. For the reduction of the feature search scope, the feature search scheme is then designed to determine redundant features based on the feature redundancy. These earlier redundant features are not considered during the later process of feature selection. Furthermore, the algorithm is proposed by combining the sample filtering mechanism and the feature search scheme. Numerical experiments finally demonstrate the effectiveness and efficiency of the proposed algorithm.