关于预测误差平方和最小准则的几点看法

A Note on The Stepwise Algorithm For Selecting Predictors on The Basis of The Criterion From Prediction Sum of Squares

在应用桃棣荣等提出的预测误差平方和（ＰＲＥＳＳ）最小准则逐步算法的过程中，发现单纯应用ＰＲＥＳＳ值来控制引进因子数目有时不很有效。因此建议：１）当引进因子过多时，采用赤池信息准则对因子数做进一步控制。２）当引进因子过少时，取消原算法中关于ＰＲＥＳＳ值上界的规定和增加候选因子数以增加引进因子数。最后。

n the stepwise algoeithm for selecting predictors on the basis of the criterion from prediction sum of squares(PRESS) proposed by Yao and Yu(1992), the criterion for controlling the number of predictor is not always efficient. In this paper the following cnackes for the issue are given. If the selected predictors are too less we increase the number of origional pedictor. Then we may get more selected predictors. If the selected predictors are more, we use the Akai Information Criterion. In additon the upper limit for the value of PRESS given by Yao and Yu should be cancelled in order to increase the number of selected predictor in some cases. Also the optimation of the selecting preditors on the basis of the criterion from prediction sum of squares is discussed.