Tell me more ×
Cross Validated is a question and answer site for statisticians, data analysts, data miners and data visualization experts. It's 100% free, no registration required.
up vote 1 down vote favorite

If one is given several hundred features (of both categorical and continuous type) what are some approaches to determining which features to keep or even drop? Data as such is difficult to visualize and other than PCA (which I think qualifies because it throws away information deemed not as important by variance) I'm not sure what other common data pruning/cleanup methods I can look into?

share|improve this question
PCA is probably not the best method to deal with a mix of continuous and categorical predictors. Are you interested in performing a separate or prior feature selection, or are you looking after modeling strategies with some form of penalization that allow to cope with large number of predictors? – chl♦ Apr 3 '12 at 22:06
If you take a quick look around you'll see about 8 questions dealing with this topic from the last few weeks alone. – rolando2 Apr 3 '12 at 23:05
I am familiar with ridge and lasso penalties to cope with a large number of predictors and have read some techniques like support vector machines are good at coping with the curse of dimensionality but I'm looking for ways to spot samples or features which may be throwing me off. With mixed data and so many features though I'm not sure of equivalents of cooks distance or metrics/ways to define "outliers" and that. – Palace Chan Apr 3 '12 at 23:12

Know someone who can answer? Share a link to this question via email, Google+, Twitter, or Facebook.

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Browse other questions tagged or ask your own question.