Abstract: Classification is a common problem in modern industrial production.The classification efficiency and classification accuracy can be improved effectively by using feature selection to filter useful information before classifying tasks.Considering the maximum correlation between features and class and the minimum redundancy among features,the minimal-redundancy-maximal-relevance(mRMR) algorithm can effectively select feature subset.However,two problems exist in the algorithm mRMR,i.e.,the relatively large deviation of the importance of the features in the middle and later stages of mRMR,and the feature subset is not given directly.A novel algorithm based on the principle of mRMR and discernibility matrix of neighborhood rough set was proposed,to solve these problems.The significance of the feature is defined by employing neighborhood entropy and neighborhood mutual entropy based on the principle of the minimal redundancy and maximal relevance,which can deal the mixed type date better.Dynamic discernibility set is defined based on the discernibility matrix.The dynamic evolution of the discernibility set is utilized as the policy to delete redundant features and narrows search range.The optimized feature subset is given when the iteration is stop by the stop condition given by discernibility matrix.In this paper,SVM,J48,KNN and MLP were selected as classi-fiers to evaluate the performance of the feature selection algorithm.The experimental results on the public datasets show that the average classification accuracy of the proposed algorithm is about 2% more than that of previous algorithm,and the proposed algorithm can effectively shorten the feature selection time on the data set with more features.Therefore,the proposed algorithm inherits the advantages of discernibility matrix and MRMR,and can effectively deal with feature selection problems.

Key words: Discernibility matrix, Feature selection, mRMR, Neighborhood rough set