2
$\begingroup$

I am trying to conduct PCA on a dataset with 17 features (which includes dummy variables; I converted two categorical variables into their corresponding dummy variables), and the first two principal components have a total explained variance ratio of 27% (15% and 12% respectively). I am trying to understand what does this mean?

  1. PCA/ dimensionality reduction is not appropriate here as I am losing out on 73% of the information?
  2. I shouldn't have included dummy variables in my PCA analysis?
  3. Converting two categorical variables to their corresponding n-1 dummies is not a good idea?

Can you please help me, I am new to PCA!

Improve this question
$\endgroup$
3
  • 1
    $\begingroup$ hopefully you have scaled your features before the PCA. The first two dimension only captures 27% of the variance, but this might be highly possible if your features are not correlated $\endgroup$
    – StupidWolf
    Commented Jan 10, 2021 at 6:36
  • $\begingroup$ it really depends on what is the purpose of doing dimension reduction. If it is to explore the data this is fine, but if it is something else you should elaborate. If you have categorical variables.. and want to analyze them as a whole, converting them to dummy is ok $\endgroup$
    – StupidWolf
    Commented Jan 10, 2021 at 6:37
  • $\begingroup$ I did scale my features using the standard scaling technique @StupidWolf. I was using PCA to find the most important features, do you recommend any other method? $\endgroup$
    – pandi20
    Commented Jan 18, 2021 at 19:43

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.