Non-normal Multivariate Linear Model Of Second Order Deviation Correction Akaike Information Criterion

information criteria for such data for variable selection has very good effect,but if the real distribution of data of non-normal distribution,often do not have a very good method for variable selection.This paper introduces a bias correction of Akaike information criterion,the rule of information to solve when the real distribution of observed data is an unknown non-normal distribution,it can provide multivariate normal linear regression model with a variable selection method with better effect.Is known to all,when the real distribution of the data of non-normal distribution,use AIC criterion for variable selection,the deviation is O(1),and there is some deviation correction method of information criteria,first order them this deviation can be reduced to O(n ^(1)),where n said sample size.This paper proposes a new rule of information,the information criterion in the AIC criterion of deviation correction on the basis of slightly made some adjustments.Although this adjustment only through constant coefficient,but this kind of new standard deviation reduced to O(n ^(2)).At the same time,the new guidelines of variance is also reduced.Finally,numerical experiments,we validate our new method of information criteria is better than before.