Does PCA apply on the raw data or the data needs to be pre-normalized before applying pca function?

Hi,

According to the link below,

https://uk.mathworks.com/help/stats/pca.html

coeff = pca(X)

I'm wondering whether this X is raw data or the raw data being normalized? Is it essential to normalize the raw data before feed into pca function?

[Adam]

It depends what you want to interpret from it. It isn't compulsory to normalise data but as with any statistical operation on multiple observations, if the ranges of the inputs are significantly different it will skew your results in that same way. The resulting principal components will be within the context of the input data you give it though so if you normalise then the principal components will reflect that and if you don't then they will likewise reflect the raw data - so you would have eigenvalues that are very skewed by larger data ranges.

[Salad Box]

Thanks for your reply. Seems there is no harm normalizing the data at the beginning. :)