Is it is possible to use cross validation to find the number of component in PCA



I have studied two methods one cross-validation and other is PCA.

cross-validation- It helps us to choose the tuning parameter of the model which increase the performance of the model on test data set.

PCA- It reduces the number of predictors into a manageable size and each component is the linear combination predicators .

There is always one problem in PCA to decide the number of components.What I want to know is it possible to use cross-validation for finding the number of principal components.


Hi @harry, read this blog for a deeper understanding of PCA


Hi @harry

as PCA is deterministic by design cross validation could only be used with the model you will use. In few words you use a wrapper to select the Principal Component as they will be your new variable, which algorithm to use to select is perhaps your question.
You can use all the one for features selection from GA, stepwise or Recursive Feature Elimination, etc …
Or simple you build a loop through your Principal Components which calculate the variance explained and decided how many components to include based on the variance explained you think is adequate.

Hope this help.