Robust Principal Component Analysis (RPCA)
Also known as: RPCA, robust principal component analysis, low-rank plus sparse decomposition, Robust Temel Bileşen Analizi (RPCA)
Robust Principal Component Analysis is a dimensionality-reduction method that extracts reliable components when the data are contaminated by outliers and noise. Introduced by Candès, Li, Ma and Wright (2011), and developed in the ROBPCA approach of Hubert, Rousseeuw and Vanden Branden (2005), it separates a data matrix into a clean low-rank part and a sparse outlier part.
Key highlights
- Extracts reliable components even when outliers and noise are present, where classical PCA would be distorted.
- Separates the genuine low-rank structure from sparse gross errors, so anomalies can be inspected directly in the sparse part.
- Well suited to high-dimensional, multivariate continuous data.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use RPCA for exploratory or descriptive analysis of high-dimensional, continuous, multivariate data that you suspect contains outliers or gross errors, with a reasonable sample size (at least about 30 observations). It is appropriate when the data matrix can plausibly be viewed as a low-rank structure plus a sparse set of corruptions. Prefer classical PCA when the data are clean or the sample is small, and treat results cautiously when the share of corrupted entries is very large.
Strengths & limitations
- Extracts reliable components even when outliers and noise are present, where classical PCA would be distorted.
- Separates the genuine low-rank structure from sparse gross errors, so anomalies can be inspected directly in the sparse part.
- Well suited to high-dimensional, multivariate continuous data.
- With small samples (n < 30) the low-rank decomposition becomes unreliable, and classical PCA is preferable.
- When the outlier ratio is very high (above about 25%), the sparse-matrix estimate becomes unstable.
- Relies on the assumption that the data are genuinely a low-rank structure plus a sparse set of corruptions.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
How is RPCA different from classical PCA?
Classical PCA fits components to all the data at once, so a few gross outliers can rotate the principal directions. RPCA assumes the data are a clean low-rank structure plus a sparse set of corruptions, pulls the corruptions into a separate matrix, and computes components from the clean part only.
What do the two matrices L and S represent?
L is the low-rank matrix that captures the genuine low-dimensional structure of the data, and S is the sparse matrix that collects the scattered outliers and gross errors. The principal components come from L, while S can be inspected to locate the anomalies.
When should I prefer ordinary PCA instead?
When the data are clean, or when the sample is small (n < 30), the low-rank decomposition becomes unreliable and classical PCA is the better choice.
What happens if there are too many outliers?
RPCA assumes the corruptions are sparse. When the outlier ratio is very high (above roughly 25%), the sparse-matrix estimate becomes unstable and the decomposition can no longer be trusted.
Sources
- 1.Candès, E. J., Li, X., Ma, Y., & Wright, J. (2011). Robust Principal Component Analysis? Journal of the ACM, 58(3), 1-37.
- 2.Hubert, M., Rousseeuw, P. J., & Vanden Branden, K. (2005). ROBPCA: A New Approach to Robust Principal Component Analysis. Technometrics, 47(1), 64-79.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 1). Robust PCA. ScholarGate. https://scholargate.app/statistics/robust-pca