考虑错分损失不同时的逐步多组线性判别和二次判别

STEPWISE MULTI-GROUP LINEAR DISCRIMINATION AND QUADRATIC DISCRIMINATION CONSIDERING DIFFERENCES OF ECONOMIC LOSSES OF INCORRECT CLASSIFICATION

目前使用的多组判别方法认为不同组间的错分的经济损失是相同的,也就是损失函数,L(h/g)=1,(h≠g)。本文研究了错分损失不同时的多组线性判别和多组二次判别。结果表明,考虑错分损失时,极端级别的判别效果可得到改善,经济损失可以明显减少。

It is regarded that the economic Losses of incorrect classification among different groups are the same, i. e. the loss function L(h/g)=l(h≠ g) in the multi-group discrimination methods which are now used. This paper studies the multi-group linear discrimination and the multi-group quadratic discrimination when the losses of incorrect classification are different. The results show that the discrimination effects of extreme classes can be improved and the economic losses can be obviously reduced when the losses of incorrect classification are considered.