Hi to everyone,
I'm trying to do a PCA analysis and then a PLS regression on a dataset composed of a lot of descriptors with diferente magnitude. I know that it's necesary scale and center the data before this kind of analysis but. Do I also have to normalize the data (with Box Cox or Yeo Johnson method) before doing this multivariate analysis and regression? If you have a paper about that I would appreciate it.
Thank you.
I am not totally sure of the nature of your question.
If you have descriptors of different natures and on which the scales are not comparable, it is compulsory to scale the data. Otherwise, the numerical value of the descriptor will play a role in the models : the large value will play a larger role in the models than the small ones. The simplest scaling I use personnally is simply the division of each variable by its standard deviation.
Kind regards,
Dominique
I agree with Dominique Bertrand . If you scale you might also wish to consider centering your variables by subtracting the mean. Best, D. Booth
Thank you all for your suggestions,
I know that for PCA analysis or PLS regression it's necessary scale and center the descriptors, but specificaly my question is if also I have to convert my data to a normal distribution using Box-Cox or Yeo-Johnson.
It is necessary that the data in the descriptors have normal distribution? Or the non-normal behavior it's not a problem in a PLS or a PCA analysis?
Hello Edward,
No, you don't have to convert your data to normal distributions. If you convert your data to normal you will probably loose some of important information within your data.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression