Research On Approaches To Quotient Spaces Feature Selection

Granular computing is a knowledge acquisition method to simplify complex problems.Fuzzy set,rough set and quotient space are three models to deal with complex problems.Fuzzy set and rough set mainly solve the problem of information uncertainty,while quotient space solves the problem of ambiguity of the same concept through multi-granularity strategy.Feature selection is the knowledge acquisition methods for selecting the most predictive features.After feature selection,the selected features retain the original meaning of the original information system.In fact,effective feature selection technology can select key features for downstream tasks,thus minimizing the time and space complexity of downstream tasks and improving the performance of learning algorithms.In this paper,a feature selection algorithm based on hierarchical quotient space is proposed.The structure of hierarchical quotient space is constructed to select the optimal granularity from the whole situation.The feature subset is selected by a new positive domain calculation method.The dependency degree is used as an evaluation index to evaluate the selection of feature subset.The pseudo code and flow chart of the algorithm are given.By comparing the results under different classifiers,the experimental results show that the proposed method is superior to other feature selection methods in feature dimension and classification accuracy.In the process of constructing hierarchical quotient space,the algorithm needs to consider global feature search for the optimal value,which will cause loss in time.Therefore,this paper proposes a method of dynamic generation of hierarchical quotient space,which reduces the complexity of time and space by constructing hierarchical quotient space layer by layer,chooses the best granularity for feature selection,and fundamentally reduces the data dimension by sampling.The pseudo-code and flow chart of the algorithm are analyzed in detail.Finally,the results of feature selection based on dynamic generation of hierarchical quotient space are compared and analyzed under different classifiers.The experimental results show that the classification accuracy of this algorithm is improved in different degrees compared with other feature selection methods.The influence of different thresholds on the accuracy of data classification is analyzed by experiments.It is proved that different sampling ratios have no significant effect on the classification accuracy of data sets.