I am trying to find out when for creating classification rule to use linear discriminant function and when to use logistic regression? I need to help to find information sources to this topic. Any clue?
Asked
Active
Viewed 1,393 times
1
-
1My advice would be first to try search `logistic discriminant` on this site. There is a dozen relevant questions already posted. – ttnphns Jan 02 '15 at 12:26
-
1... with http://stats.stackexchange.com/questions/14697/are-fishers-linear-discriminant-and-logistic-regression-classifier-related having the most complete answers. Note there that since logistic regression is derivable from LDA by Bayes' rule, if the assumption-laden LDA fits the data, the logistic necessarily does also. So there is little reason to use LDA. – Frank Harrell Jan 02 '15 at 13:48
-
1See http:biostat.mc.vanderbilt.edu/wiki/pub/Main/FrankHarrell/discrim.pdf for my paper explaining why Efron's result is misleading. – Frank Harrell Jan 04 '15 at 03:41
1 Answers
0
If you have mix of continous and categorical predictors you can use logistic regression.
If you have only continuous predictors then Discriminant analysis can be used because Discriminant analysis can't have categorical predictors.
Discriminant analysis can give better accuracy than logistics regression but it varies with the data and context etc.

GeorgeOfTheRF
- 5,063
- 14
- 42
- 51
-
1You need to also mention that the continuous predictors in LDA need to be normally distributed. – gung - Reinstate Monica Jan 04 '15 at 03:04