基于改进邻域粒的模糊熵特征选择算法

Fuzzy entropy feature selection algorithm based on improved neighborhood granule

特征选择是一项重要的数据预处理技术,其目的是在不降低数据分类精度情形下选择一个特征子集,从而对原数据集达到降维的效果,同时也提高学习算法的性能.在邻域粗糙集模型中,传统方法构造出的对象邻域粒未考虑数据的分布问题,使得邻域粒存在一定的误差.首先通过方差来刻画数据的分布,然后根据数据分布提出一种改进的邻域粒,这种改进的邻域粒能够自适应数据的分布,有着较好的优越性,最后将改进邻域粒与邻域模糊熵结合,提出一种特征重要度的评估方式,并给出对应的特征选择算法.实验结果表明,新提出的特征选择算法在特征选择结果、时间消耗和特征子集的分类精度方面都更具一定的优越性.

Feature selection is a significant technology of data preprocessing in areas of machine learning and data mining,and its purpose is to select a subset of features without reducing the accuracy of the classification of data,and achieving the effect of dimension reduction for the original data set.At the same time,the performance of learning algorithm is improved as well.In neighborhood rough set model,the similarity relationship of the between objects is described by neighborhood granule.However,the neighborhood granule is constructed through the traditional methods without consideration of the problems about the data distribution,and it makes some errors exist in the neighborhood granule.In this paper,the data distribution is firstly described through the variance.Then,an improved neighborhood granule is proposed according to the data distribution.The improved neighborhood granule has an a-daptive data distribution.In addition,the improved neighborhood granule has more superiority compared with the traditional neighborhood granule.And then,on the basis of the improved neighborhood granule,combing the improved neighborhood granule with neighborhood fuzzy entropy,making the features in the information system have better evaluation of importance.At last,the algorithm of feature selection based on fuzzy entropy is proposed according to the evaluation of importance,which is the better importance evaluation through the combination of the neighborhood granule and the neighborhood fuzzy entropy.The experimental results show that the new proposed feature selection algorithm can select smaller feature subset.At the same time,the feature subset keeps more accuracy of the classification.In addition,the proposed algorithm of feature selection also has higher efficiency.Therefore,the proposed feature selection algorithm has more superiority.