Research On Grouped Sorting Feature Selection Method Inspired By Visual Attention Mechanism

In the era of big data,in order to obtain the effective information fast from the large scale high-dimensional data,feature selection is of vital importance and has attracted a lot of attention.Owing to the advantage of maximizing the correlation degree between features and labels in it while taking into account minimizing redundancy between the features,the feature selection method based on feature grouping has been widely accepted.While the saliency based calculation form and specificity based processing mechanism of visual attention mechanism produce a wonderful inspiration on the grouped sorting feature selection method.Inspired by this mechanism,the specificity based processing strategy for correlated information and saliency based calculation strategy for significant information were simulated to form a new ideas based on grouped sorting in this thesis,and Grouped Sorting Feature Selection(GSFS)algorithm was proposed.Firstly,started with the similarity of the feature grouping and sorting process together with the above two strategies of the visual attention mechanism,the simulation—inspiration relationship between GSFS algorithm and visual attention mechanism was introduced.Combined with the introduction of the feature grouping indexes and sorting indexes,GSFS algorithm based on maximum information compression index and Fisher score was proposed and its basic principle was introduced in detail.The suitable criteria to ascertain the best feature group number and the optimal feature subset were constructed and the process procedure of the algorithm was realized.Then the classification experiments and redundancy removing effect evaluation on 8 standard data sets with different dimensions and different number of classes compared with six kinds of classical feature selection methods were carried out.The results show that the optimal feature subsets selected by GSFS have excellent classification ability and contain least redundant information.The effectiveness of the GSFS algorithm to solve the feature selection problem is validated.On the basis of the theory of GSFS algorithm,the Distributed Processing based Grouped Sorting Feature Selection(DP_GSFS)algorithm was introduced in this thesis to solve the problem of high computational complexity in high-dimensional data sets feature selection further.It used the method of equally dividing the original feature set to reduce the computational complexity of the correlation measurement between feature pairs.On the basis of the fact that its classification ability and the redundancy removing effect have been validated by the classification experiments and redundancy evaluation on 6 standard data sets,the results of computational efficiency experiments on those two methods show that the DP_GSFS method can significantly reduce the computational time when setting the appropriate number of groups,which validates its ability to improve computational efficiency.And it also can provide valuable reference for the fast analysis and processing of large scale high-dimensional data sets in the future.In addition,the algorithms were applied to the real medical image feature data sets.The feature selection effect and classification results demonstrate their ability to solve practical problems.