Principal Component Analysis (PCA) vs t-SNE
THE VERDICT
They answer different questions. PCA is a fast, deterministic, invertible linear projection - use it for compression, denoising, and as model input. t-SNE is a non-linear microscope for 2-D visualisation: clusters pop beautifully, but distances and axes stop meaning anything, and there is no transform for new data. Pipeline them: PCA to 50 dimensions, then t-SNE for the picture - and never feed t-SNE output to a model.
[ 01 ] Side by side
| Dimension | Principal Component Analysis (PCA) | t-SNE |
|---|---|---|
| Family | Dimensionality Reduction | Dimensionality Reduction |
| Interpretability | Medium | Low |
| Training speed | Fast | Slow |
| Data needed | Medium | Medium |
| Complexity | Low | Medium |
| Training cost | O(n·d²) or O(n·d·k) with truncated SVD | O(n log n) with Barnes-Hut; practical ceiling ~50-100k points |
| Inference cost | O(d·k) | no natural transform for new points |
[ 02 ] When to choose each
Choose Principal Component Analysis (PCA) when…
- Data visualization
- Feature reduction
- Noise filtering
…but not when
- The signal lives on a curved manifold (use UMAP/t-SNE/autoencoders for that)
- You need the reduced features to stay individually meaningful to stakeholders
- Variance ≠ importance for your task: low-variance directions can carry the label signal
How it works: Rotate the axes to point along the directions where the data actually varies, ordered by how much variance each direction carries. Keep the first few, drop the rest, and you have compressed the data with the least possible information loss (in the linear, squared-error sense). One rotation - no learning loop, no local minima.
Full Principal Component Analysis (PCA) dossier →Choose t-SNE when…
- 2D/3D visualization
- Cluster exploration
- Embedding visualization
…but not when
- As input features for downstream models - it distorts distances by design
- When between-cluster distances or axis directions need to mean something
- Datasets past ~100k points (use UMAP) or pipelines needing transform() on new data
How it works: Make a 2-D map where points that were close neighbours in high dimensions stay close. t-SNE converts distances to neighbour probabilities in both spaces and drags the map around until the two distributions agree. It is a microscope for local structure - clusters pop beautifully - but the map’s global geometry is essentially fiction.
Full t-SNE dossier →[ 03 ] Quick answers
Q.01When should I use Principal Component Analysis (PCA) instead of t-SNE?
Principal Component Analysis (PCA) is the better choice for: Data visualization; Feature reduction; Noise filtering. Avoid it when: The signal lives on a curved manifold (use UMAP/t-SNE/autoencoders for that)
Q.02When should I use t-SNE instead of Principal Component Analysis (PCA)?
t-SNE is the better choice for: 2D/3D visualization; Cluster exploration; Embedding visualization. Avoid it when: As input features for downstream models - it distorts distances by design
Q.03Is Principal Component Analysis (PCA) or t-SNE easier to interpret?
Principal Component Analysis (PCA): medium interpretability. t-SNE: low interpretability. They answer different questions.