Every feature has unit variance, yet features remain correlated. What extra operation would whitening add?
Classic MLLevel: IntermediateDecorrelation
Every feature has unit variance, yet features remain correlated. What extra operation would whitening add?
Hints
- A diagonal rescaling cannot generally remove correlation.
- Consider rotating to principal axes.
Answer
Independent scaling does not rotate coordinates. PCA whitening also decorrelates retained components and rescales their variance, potentially discarding useful relative-variance information.
Source and licence: scikit-learn 1.7.2 — preprocessing · BSD-3-Clause · answer: Quantoros editorial