0

Y is the dependent variable with the outcome 0 and 1, and so are X1...X140. As far as I know, I can't use the simple lasso regression in order to look at which variables are shrinked down, since we got only binary data.

We also got perfect linear independence, where for example X1 stands for male and X2 for female. Furthermore, I want to do a cross validation, in order to determine which lambda value is the best.

I was reading, that some people suggested to use grouped lasso regression, but wouldn't a logistic lasso regression be more appropriate? I'm using glmnetand the caret packages, but I couldn't find in the documentation, which solves my problem. Which packages do you know, who can deal with those problems?

Textime
  • 55
  • 6
  • Are you asking if you should predict binary variable with (penalized) linear vs logistic regression? – Tim May 08 '19 at 10:56
  • It's more (penalized) linear vs (penalized) logistic. Maybe you meant that with your question. And again, **all** variables are binary – Textime May 08 '19 at 11:07
  • Still, the outcome is binary, so why linear regression should be better? Regression (logistic, linear, any) handles binary features. – Tim May 08 '19 at 11:23
  • I marked your question of a duplicate question on regression. The fact that you add the lasso penalty to the loss term doesn't matter. – Tim May 09 '19 at 06:59

0 Answers0