Simulation And Empirical Analysis Of GDP Based On Lasso And Its Improved Method

With the advent of the era of large data artificial intelligence,the traditional regression analysis method has failed in large number of high-dimensional and independent variables with multiple collinearity.In view of the limitations of regression model and the shortcomings of subset feature selection method and ridge regression feature selection method,Lasso feature selection method and its improved algorithm are proposed,and Lasso algorithm is simulated and analyzed empirically in terms of GDP.When multiple collinearity occurs between independent variables,the least squares estimates will be unstable,variance will increase,and the prediction accuracy of regression model is very low.Therefore,how to solve the problem of multiple collinearity with many independent variables becomes the key.Subset selection method and ridge regression method can be used to deal with this kind of problem,but both subset selection method and ridge regression method have shortcomings.Subset selection method is a discrete variable selection process,either selecting variables or eliminating variables,so important information may be lost.Ridge regression method is a continuous variable compression process,and does not shrink any coefficients to 0,ridge regression method can not give a simple and explainable model.Lasso method combines the advantages of these two methods.This paper mainly introduces the definition of Lasso regression and the corresponding algorithm of lasso regression.It also introduces the coordinate axis descent algorithm and the minimum angle regression algorithm.Because of the superiority of the minimum angle regression algorithm,this paper mainly introduces the algorithm and its two guiding algorithms,forward selection algorithm and forward gradient algorithm.In view of the shortcomings of Lasso regression method,two improved methods of lasso regression model are introduced,which are adaptive Lasso model and Elastic Net model.The simulation experiment of minimum angle regression algorithm is carried out to further understand the process of variable selection of the algorithm.Finally,the minimum angle regression model,the Adaptive Lasso model and the Elastic Net model are established to carry out empirical analysis of GDP.By comparing these three models,it is concluded that the Adaptive Lasso model has the minimum prediction error,and some suggestions are made according to the regression coefficient of the model.