Research And Application Of Model Selection Method Based On Information Criterions

The correct handling of data information has become the requirements of this era, the reason is very clear, that is the arrival of the big data era. So that all aspects of human life are full of data. In addition, with the development of modeling theory, almost all data can be found to be suitable for some model. However, the accuracy of data depends on the particular model structures, so the best model selection method has very important significance in the statistical analysis. An important basis for the selection of the best models for various statistical analysis and many scientific research is the model selection criterion.Multivariate multiple linear regression model contacts linear regression model with variable variables, which makes it become a major development of statistical methods. It is applicable to every social domain. At the same time, information criterion, such as AIC, BIC and FIC criterion, is a very important discovery in the field of statistics. Their appearance is a good solution to the problem of model selection. However, because of the information criterion formula is simple and general, which in some special models, such as the sample size is too small, too many independent variables in the model, the performance is not ideal.And these models in some fields maybe play a decisive role. So statisticians hope to be able to make corrections to the information criterions for these special models.This paper focuses on the AIC criterion of multivariate multiple linear regression model,and making a correction. The corrected AIC criterion can better optimize the model, and the simulation research is carried out to prove the accuracy of the corrected AIC criterion. In the end, this corrected criterion is being applied to an instance of the life.