If 12 covariates of data are all count data (looks like a Poisson dist with the highest peak at 0), what are reasonable pre-processing methods that might make PCA more effective?

  • 1
    $\begingroup$ Welcome to SE.DataScience! The "do" and "don't" heavily depend on what you mean by "more effective"? If it is not clear, I suggest going without any pre-processing until goals, and consequently flaws become more clear. $\endgroup$ – Esmailian Apr 9 at 19:05
  • $\begingroup$ Well, I'm extracting my loadings from the PCA model to use elsewhere as a kind of weight. But given that the results of PCA differ greatly depending on how one standardizes or preprocesses their data beforehand, I'm just trying to prevent high counts from totally skewing my data. I've tried just removing obvious outliers (and a huge mass of rows with all 0s) but the data is still heavily leaning toward a poisson which is making my score plots really wacky and concerning. My overall goal is to maximize the spread of my variation. $\endgroup$ – ag_ojo Apr 9 at 20:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.