How does PCA interact with feature normalization techniques beyond standard scaling?

PCA is sensitive to feature normalization, and different scaling methods can significantly alter component directions.

PCA relies on variance structure, so preprocessing choices strongly influence results. StandardScaler enforces zero mean and unit variance, making all features equally weighted. RobustScaler reduces impact of outliers, while MinMaxScaler compresses ranges but may distort variance relationships. Whitening further transforms PCA outputs into uncorrelated unit-variance components. Each normalization method changes covariance structure and therefore eigen decomposition results.