Timeline for using random forest for missing data imputation in categorical variables ( in R)
Current License: CC BY-SA 3.0
11 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jul 11, 2014 at 16:13 | comment | added | dfife | You might want to try this approach: stats.stackexchange.com/questions/49270/…. It seems reasonable (first iteration=median imputation, second iteration = impute with proximity, repeat). | |
| Jul 10, 2014 at 19:11 | comment | added | John | The purpose is to complete the data so that i can do other analysis that requires complete data. one thing I can simply do is calculate mode and replace missing value with mode (most frequent value) but I think more powerful approaches do exist and want to take advantage of them. | |
| Jul 10, 2014 at 15:37 | comment | added | dfife | I'm still not quite able to see the purpose. Are you trying to do an unsupervised clustering the data? (That's the only explanation for why you wouldn't have a dv that I can think of). Maybe a bit more explanation on the substantive question you're trying to address (e.g., "each of the variables are binary responses to a political questionnaire and we're trying to segregate people into clusters based on their responses.") | |
| Jul 10, 2014 at 0:33 | comment | added | John | please see my recent edits. | |
| Jul 9, 2014 at 19:13 | comment | added | dfife | You mentioned you were trying to predict...what are you trying to predict if there's no y? What is the end result? Perhaps a little more context would help. | |
| Jul 9, 2014 at 18:23 | comment | added | John |
my question here is : None of the above variables are y variables, but each can be used as y, but the problem here is that every variables have at least NA making them unsuitable for using rfImpute
|
|
| Jul 9, 2014 at 15:03 | comment | added | dfife | You set up the rfImpute exactly as you do for the randomForest part. See ?rfImpute for an example using the Iris data. [iris.imputed <- rfImpute(Species ~ ., iris.na); iris.rf <- randomForest(Species ~ ., iris.imputed)] | |
| S Jul 9, 2014 at 14:59 | history | suggested | John | CC BY-SA 3.0 |
formatting codes and text to make clear
|
| Jul 9, 2014 at 14:54 | comment | added | John | I am interested in imputing - the first point you made. Double checking if this works for categorical variables - what would be y and x variables ? | |
| Jul 9, 2014 at 14:34 | review | Suggested edits | |||
| S Jul 9, 2014 at 14:59 | |||||
| Jul 9, 2014 at 14:12 | history | answered | dfife | CC BY-SA 3.0 |