f**********t 发帖数: 1001 | 1 1. 想了解一下公司里做这个的时候,是以什么标准来评判一个credit score model的
好坏
的?
2. 另外,一个公司里是不是会有多个credit score model呢?这个问题还望牛人指教
。看相
关的书了解到的情况,是说对历史数据做个logistic regression,这样对于每个
factor都会
有相应的coefficient。然后新的数据到来时,已知coefficient,就能做预测。想问的
是,这
多个credit score model是如何产生的?是不是不同的binning method(相似的组归类),
factor的取舍,都会产生不同的model?
多谢~~~~ |
s*******e 发帖数: 1385 | 2 帮东岸的牛mm友情的顶一下
类),
【在 f**********t 的大作中提到】 : 1. 想了解一下公司里做这个的时候,是以什么标准来评判一个credit score model的 : 好坏 : 的? : 2. 另外,一个公司里是不是会有多个credit score model呢?这个问题还望牛人指教 : 。看相 : 关的书了解到的情况,是说对历史数据做个logistic regression,这样对于每个 : factor都会 : 有相应的coefficient。然后新的数据到来时,已知coefficient,就能做预测。想问的 : 是,这 : 多个credit score model是如何产生的?是不是不同的binning method(相似的组归类),
|
l****u 发帖数: 529 | 3 Only if the target variable is 1 or 0 can the predictive model be
logistic regression. If the target variable consists of a sequence of
numbers, such as people's real scores, decision tree will be a better way
to go. |
A*******s 发帖数: 3942 | 4 there is no "real scores". credit score is the log transformation of the
predicted probability from logistic regression.
【在 l****u 的大作中提到】 : Only if the target variable is 1 or 0 can the predictive model be : logistic regression. If the target variable consists of a sequence of : numbers, such as people's real scores, decision tree will be a better way : to go.
|
s*********e 发帖数: 1051 | 5 are you sure?
【在 A*******s 的大作中提到】 : there is no "real scores". credit score is the log transformation of the : predicted probability from logistic regression.
|
A*******s 发帖数: 3942 | 6 if u know sth, then say it.
【在 s*********e 的大作中提到】 : are you sure?
|
f**********t 发帖数: 1001 | 7 汗,不要听Alex童鞋乱说。他都不认识我,哈哈
【在 s*******e 的大作中提到】 : 帮东岸的牛mm友情的顶一下 : : 类),
|
l****u 发帖数: 529 | 8 Thanks Actuaries. Good to know that. Then it will be much easier.
Another question, how to get the credit scores? It seems that only log
transforming the probability is not enough.
【在 A*******s 的大作中提到】 : there is no "real scores". credit score is the log transformation of the : predicted probability from logistic regression.
|
f**********t 发帖数: 1001 | 9 Thanks for answering.
不过公司里面怎样评判一个模型是好还是坏呢?
way
【在 l****u 的大作中提到】 : Only if the target variable is 1 or 0 can the predictive model be : logistic regression. If the target variable consists of a sequence of : numbers, such as people's real scores, decision tree will be a better way : to go.
|
f**********t 发帖数: 1001 | 10 My 2 cents:
According to my understanding, the credit scores can have multiple
levels, say 1-10. And we can have a map function, if the probability is
within an interval, then we map it to a certain level. But I don't quite
understand how to determine the exact interval.
So let's wait for big cows~~~~
【在 l****u 的大作中提到】 : Thanks Actuaries. Good to know that. Then it will be much easier. : Another question, how to get the credit scores? It seems that only log : transforming the probability is not enough.
|
l****u 发帖数: 529 | 11 What I have is only the knowledge from books, some daniu may give you more
accurate answers.
For decision prediction, the performance is evaluated by accuracy,
misclassification,or KS.
For ranking prediction, two measures of model fit can be used, ROC index and
Gini coefficient.
|
s*********e 发帖数: 1051 | 12 right on.
it is correct that the score is a monotonic function of predicted
probability.
however, there are many things to consider as well, including points to
double
the odds, reference score, and so on.
【在 l****u 的大作中提到】 : What I have is only the knowledge from books, some daniu may give you more : accurate answers. : For decision prediction, the performance is evaluated by accuracy, : misclassification,or KS. : For ranking prediction, two measures of model fit can be used, ROC index and : Gini coefficient. :
|
D******n 发帖数: 2836 | 13 never heard of this.
it is either the odds level or probability level usually accompanied with re
scaling.
【在 A*******s 的大作中提到】 : there is no "real scores". credit score is the log transformation of the : predicted probability from logistic regression.
|
A*******s 发帖数: 3942 | 14 ok my bad. i was going to say linear transformation, but i realized it's not
linear coz there is a logit in it. so i say log...
anyway, hard to describe it using a few words. maybe linear transformation
of log odds.
re
【在 D******n 的大作中提到】 : never heard of this. : it is either the odds level or probability level usually accompanied with re : scaling.
|