A New Method Of Repetitive Screening For Regression Analysis Of Multivariate Supersaturated Models

Multivariate linear regression is an important analytical method and used widely. However, there are two difficult problems in it. One is multicollinearity, the other is oversaturated model. The commonly used methods have been showed as principal components regression, ridge regression, optimal subset method and screen stepwise regression can not solve these two problems well.My research is aimed at improving the screen stepwise regression to get a good method of multivariate linear regression.This study is based on matlab platform programming, using simulated data for testing the merits of the program and improving continuously. In addition, I have this algorithm and Association analysis to compare. At last, I analyze the actual datas with the program, and further verify the validity and reliability of the method at the same time.Analysis show that my results are close to the true values, and sensitivity and specificity are high. And the improved calculation efficiency also has no small increase. The new algorithm is better than the original method in both the effect and the efficiency..