0

I had a dataset that was terribly haunted by colinearity. I carried out PCA on it to remove the colinearity. Let's assume I am going to build a linear regression model, which is easily explainable, on the PCA'd data. Since the data columns are no longer the original ones, how can we find the most important one original feature according to the model, or say, explain the model using the original input features?

Memphis Meng
  • 140
  • 4
  • 1
    But each PC is a linear combination of original variables. See https://stats.stackexchange.com/q/126885/3277 – ttnphns Feb 26 '21 at 06:50
  • 1
    You would need to define what you exactly mean with "feature importance". Such definition is essential in answering this question. As @ttnphns say, you can map the estimated regression equation to the original variables. But this won't tell you much about their importance due to collinearity etc. – Michael M Feb 26 '21 at 06:52
  • Agreed with @MichaelM, PCA more or less affects the explanability of my model. I heard that Factor Analysis is good at telling stories. Is it possible to tell me how it makes it? – Memphis Meng Feb 26 '21 at 22:45

0 Answers0