Calculation Results
Calinski-Harabasz score—
Between dispersion—
Within dispersion—
Detected clusters—
Valid samples—
Between mean square—
Within mean square—
Feature count—
Cluster Details
| Cluster | Size | Centroid | Within dispersion |
|---|
Configuration Comparison
| Rank | Configuration | Clusters | Score | Within | Between |
|---|
Score by Configuration
Cluster Projection
PCA is used when more than two features exist.
Cluster Sizes
Calculation Summary
Calculator
Formula Used
CH = [Bk ÷ (k − 1)] ÷ [Wk ÷ (n − k)]
Bk measures between-cluster dispersion. Wk measures within-cluster dispersion. Larger values usually indicate clearer clustering structure.
How to Use
- Paste numeric feature data with cluster labels.
- Select the label column and parsing options.
- Choose scaling and optional K-Means comparisons.
- Press calculate to inspect scores and charts.
- Export the completed results when required.
Example Data
| Feature 1 | Feature 2 | Cluster |
|---|---|---|
| 1.0 | 1.1 | A |
| 1.2 | 0.8 | A |
| 5.0 | 5.2 | B |
| 5.3 | 4.9 | B |
| 9.1 | 1.2 | C |
| 8.8 | 0.9 | C |
Frequently Asked Questions
What does a higher score mean?
A higher score usually signals compact, well-separated clusters.
Does the score have a fixed maximum?
No universal maximum exists for every dataset and model.
Can different datasets be compared directly?
Compare models using the same samples and feature preparation.
Why should features be scaled?
Scaling prevents large-unit features from dominating Euclidean distances.
What cluster counts are valid?
Use at least two clusters and fewer clusters than samples.
Can singleton clusters be used?
They are permitted, but they may distort model comparisons.
Does this calculator support PCA?
Yes. PCA projects high-dimensional results onto two chart axes.
Can external model labels be compared?
Yes. Paste additional label sets using the required format.
Is the best score always the best model?
No. Domain meaning and cluster stability also matter.