h***t
发帖数: 2540 | 1 请问对于binary classification,如果positive observation很少，也就是整个数据非
常unbalanced, 有没有比较好的建模方法？ | f*******n
发帖数: 2665 | 2 oversampling
【在 h***t 的大作中提到】
: 请问对于binary classification,如果positive observation很少，也就是整个数据非
: 常unbalanced, 有没有比较好的建模方法？
| z**********i
发帖数: 12276 | 3 zero inflated negative binomial.
【在 h***t 的大作中提到】
: 请问对于binary classification,如果positive observation很少，也就是整个数据非
: 常unbalanced, 有没有比较好的建模方法？
| D**u
发帖数: 288 | 4 1st choice:
oversampling/stratified sampling, add weight to the observations, both sas
and R can do this easily.
2nd choice:
rather simple negative binomial model than any zero inflated models for the
ease of avioding mixture model
of course, you can always try zero inflated models
a good explanation is here:
http://www.statisticalhorizons.com/zero-inflated-models | s*********e
发帖数: 1051 | 5 对binary classification, zero inflated nega. binomial根本不合适。
the
【在 D**u 的大作中提到】
: 1st choice:
: oversampling/stratified sampling, add weight to the observations, both sas
: and R can do this easily.
: 2nd choice:
: rather simple negative binomial model than any zero inflated models for the
: ease of avioding mixture model
: of course, you can always try zero inflated models
: a good explanation is here:
: http://www.statisticalhorizons.com/zero-inflated-models
| D**u
发帖数: 288 | 6 Agree, forgot it is "binary"
【在 s*********e 的大作中提到】
: 对binary classification, zero inflated nega. binomial根本不合适。
:
: the
|
|
