Research Of Feature Selection Algorithm Based On Sparse Representation

In the pattern recognition disciplines,the feature selection as an important direction within its scope,which has evolved into a hotspot in recent years.In real life,the results of scientific research have penetrated into many industries,and obtain practical application in the industries.In disciplinary research and real-life applications,we will face and deal with huge amounts of data.However,these data have a small number of samples,and its data dimension is large,at the same time,with same redundant features,which is a big challenge for the computer processing resources and processing real-time.To solve the "dimension of disaster" problem has a very important role.Therefore,feature selection plays an important role as an important step in data processing.Because of the large dimension,the regression problem of high dimensional data is a relatively large challenge.An effective solution is the feature selection.While linear regression based on sparse representation has proven to be very effective in dealing with high dimensional data.The traditional sparse representation of the linear regression algorithm is Lasso algorithm.Through minimizing the objective function,with the absolute value of the coefficient as the compression model coefficients,making the absolute value of the smaller coefficients are compressed to 0,so Lasso algorithm can remove many unless features.Due to the advantages of Lasso in the feature selection method,it has been widely recognized and used.In order to solve the feature selection problem faced by high-dimensional data,this paper is based on the sparse representation of the linear regression model,to do a further study.Based on the Lasso linear regression model,a feature selection model with discriminant information is proposed.The characteristic variables and feature variables have little repetition,and the characteristic variables have a great correlation with the response variables.At the same time,a feature selection method based on Lasso model is proposed.The selected feature reflects the higher order interaction information of the covariate and response variables.This article is tested on multiple open datasets.It can be seen from the experimental test results that the proposed model has improved the classification accuracy of the feature selection task.