midPCA
What are limitations of PCA in nonlinear datasets?
Updated May 17, 2026
Short answer
PCA cannot capture nonlinear relationships in data.
Deep explanation
PCA assumes linear transformations and variance-based structure. Nonlinear manifolds cannot be properly represented, leading to information loss. Methods like kernel PCA or t-SNE are used instead.
Real-world example
Image manifolds in computer vision datasets.
Common mistakes
- Applying PCA to nonlinear embeddings blindly.
Follow-up questions
- What replaces PCA for nonlinear data?
- Is PCA ever nonlinear?