Formula Used
Eigenvalues measure variance captured by each principal component. Larger values indicate stronger components. The ratios show each component’s relative importance.
The cumulative ratio measures retained information after selecting components. A higher target preserves more variance. It also reduces compression and computational savings.
How to Use
Select raw data, eigenvalues, variance ratios, or a matrix. Configure preprocessing and the PCA matrix. Then choose your preferred component-selection rule.
Press the calculation button to generate recommendations and diagnostics. Review agreement between different selection methods. Export the results when your decision is complete.
Example Data
| Feature A | Feature B | Feature C | Feature D |
|---|---|---|---|
| 2.5 | 2.4 | 1.2 | 0.8 |
| 0.5 | 0.7 | 0.3 | 0.2 |
| 2.2 | 2.9 | 1.1 | 0.9 |
| 1.9 | 2.2 | 0.9 | 0.7 |
Frequently Asked Questions
What does PCA component selection determine?
It determines how many principal components should represent the data. The selected components retain important variance. Remaining dimensions are removed to simplify modelling.
Which explained variance threshold should I use?
Ninety or ninety-five percent works for many exploratory tasks. Higher thresholds preserve more information. Lower thresholds create stronger compression and faster models.
When is the Kaiser criterion appropriate?
The Kaiser rule keeps eigenvalues greater than one. It is mainly appropriate for correlation-based PCA. Standardised features make its threshold meaningful.
Why can selection methods disagree?
Each method evaluates a different property of the spectrum. Variance thresholds prioritise retention. Elbow and simulation methods emphasise structure beyond noise.
Should I use covariance or correlation PCA?
Use correlation PCA when feature units or scales differ. Covariance PCA preserves original scale differences. Domain knowledge should guide the final choice.
What is parallel analysis?
Parallel analysis compares observed eigenvalues with random-data eigenvalues. Components exceeding random references are retained. This often provides a stronger statistical benchmark.
What does reconstruction error mean?
Reconstruction error measures information lost after reducing dimensions. Smaller errors indicate better approximation. More selected components normally reduce this error.
Can PCA handle missing values directly?
Standard PCA requires complete numerical observations. This calculator can remove or impute missing values. Review imputation effects before modelling important datasets.
Does high retained variance guarantee good predictions?
No, PCA is unsupervised and ignores target labels. Low-variance features may still predict outcomes. Validate selected components within your final model.