Application Of Lasso And Improved Lasso Method In Several Kinds Of Model Variables Selection

In recent years,with the rise of the era of large data,the analysis and processing of data have been paid more and more attention in the fields of social science,information science,biology,medicine and finance.And the extraction of the essential features of data and pattern discovery have become an important research direction.In the face of massive data,we need to establish a suitable mathematical model and dig out as few and fully effective data as possible to analyze it and apply it to real life and work.The lasso method is a variable selection method that can effectively handle high dimensional data and improve the accuracy of the model.It belongs to a typical coefficient compression regression method,that is,by adding the absolute value function of the model coefficients in the optimal objective function as a penalty,the coefficients of the smaller absolute values are compressed to 0,and then the selection of variables and the estimation of the corresponding parameters are realized.Compared with the traditional model selection method,lasso and the improved lasso method can overcome some shortcomings in the selection model.The lasso method and its improved method have also been paid great attention to the statistical theory and the application of various models.This paper mainly studies the application of lasso and improved lasso method in the selection of variables on the BP neural network model,the balanced vertical data model and the semi parametric Logistic model.In the first part,the lasso method and BP neural network are combined to build a forecasting model of housing demand in Guangxi.By comparing and predicting the results of the BP neural network,the BP neural network based on principal component analysis and the BP neural network based on the lasso method,it shows that the BP neural network based on the lasso method can obtain prediction effect better.In the second part,the properties of the improved lasso method in the balanced longitudinal data model are discussed,and its variable selection problem is studied.The results of the adaptive elastic network method are compared with the results of Lasso,adaptive lasso and elastic network method by numerical simulation.The preliminary numerical results show that the results obtained by the improved lasso method are more accurate.In the third part,the variable selection problem of the semi parametric Logistic model based on the improved lasso method is mainly studied.The estimation method of unknown parameters and unknown functions in the model is given,and its estimation problem is discussed.Finally,three different variable selection methods are compared by using the lasso method,the elastic network method and the adaptive elastic network method,to show that the adaptive elastic network method has a good variable selection effect.