0

I am trying on a regression model with 44 variables. Executing PCA due to multicollinearity, I receive 6 principal components. I am using Leave-one-out cross-validation.

Unfortuneately, due to PCA and its "Measure of sample adequacy", which should be >0.5 for all variables in PCA, I have to exclude some variables.

Is this the variable/feature selection part of PCA? Is it correct to say, that those variables aren't required to make a good prediction from these variables?

Thanks.

meg1234
  • 25
  • 4
  • Why only 6 PCs from 44 variables? – David Z May 23 '21 at 13:08
  • I excluded 13 variables due to a MSA of below 0.5. With the rest of the variables, I get 6 PCs due to the KAISER-criterion (eigenvalue higher than 1). On the basis of the Velicer MAP, I would need to take 6 PCs. – meg1234 May 24 '21 at 01:24

0 Answers0