Is it possible to index which variables to extract after Principle Component Analysis?

Hi All,

I am a little confused as to how I can determine which variables I want after running my data matrix X through the pca funtion.

[coeff,score,latent,tsquared,explained,mu] = pca(X);

I have a 1000x16 dataset. I know from the 'explained' output that 95% of my feature variability can be explained by 7 variables.

Is there a way of determining which 7 variables (which columns numbers from the original X matrix) these are through indexing from the original X matrix,coeff, scores matrix?

Any help would be greatly appreciated here.

Answer 1

[the cyclist]

It will not be 7 of your original variable that explain 95% of the variation. This is an important concept to understand.

Rather, it will be 7 of the new variables, each of which is a linear combination of your original variables, that will explain the variation. Each column of coeff is the weight of the original variable. The first 7 columns of coeff are what you want.

[sh10101]

Thanks for clearing this up!

So are the first 7 columns of coeff what I can now build my model using?