Initial steps with logistic regression I have a dichotomous variable which I am trying  to predict. In order to do this I have about 2000 variables out of which 1100 are continuous variables (real values between 0.0 and 1.0) and the rest are categorical variable (0 or 1).
I am a total newbie and would like to know how to go about it. I am not aware of the relationships between these variables.
Roughly the following are the steps that I am planning to take:


*

*Should I try to use correlation analysis between each variable and the target variable?

*Should I try some kind of clustering and remove some outliers and try to come up with a model.

*What other steps should I take to identify relationships between variables and come up with a model?


Thanks
 A: It would help if we knew what kind of data you're working with. And depending on the degree of correlation between your predictors (among themselves), the solution might be different.
But generally, here it is: for such a great number of predictors, what I would probably do first is to run correlations as you mention (looping through the variables, using R for instance) to eliminate the uninformative variables. Note that there is some risk in doing that; some poorly correlated variable could have an effect which is masked by a confounder. But having so many variables, it probably makes sense to eliminate a first bunch of variables that way. 
Then, after having selected your variables having a sufficient amount of correlation with the outcome, you could look into the package called BMA. It provides a function called bic.glm which will select the best sets of predictors for you. It does a little less well for categorical predictors than for continuous ones, but it would rapidly get you somewhere. 
Ideally, you would just throw the full equation and the function would look for the best predictors. But with 2000 variables, I don't think that's a good idea! :)
There are several tutorials to get started with R, if you're new to it.
Good luck!
http://cran.r-project.org/
