How to test a PCA+classifier model?

Asked Oct 08 '18 at 11:24

Active Oct 08 '18 at 11:24

Viewed 27 times

I have a 100x45 dataset and I'd like to perform feature selection and classification/regression.

I'm currently using various techniques to check which one has the best performances, but I have a doubt about how to treat PCA + SVM or random Forest or CART. With other feature selection methods and elastic net, I'm performing Cross Validation (CV) to select the hyperparameters and to test the model, since I have few records and at the moment I'm not in the condition to increase them.

With PCA which is the best course of action? Train-test or CV? Should I perform PCA on the train set and then use the number of PCs on the test set? In the case of CV how can I proceed?

asked Oct 08 '18 at 11:24

schrodingercat

Is the 100x45 a "100 rows or samples" by "45 columns or dimensions"? – EngrStudent Oct 08 '18 at 13:42
@EngrStudent yes exactly. 100 subject x 45 variables – schrodingercat Oct 08 '18 at 13:49
@schrodingercat - I like the "Boruta" with two loops, to reduce the columns to "important only". I don't know if that approach is consistent with your question. – EngrStudent Oct 08 '18 at 14:34

How to test a PCA+classifier model?

0 Answers0