Calinski-Harabasz Score Measure

A Calinski-Harabasz Score Measure is an internal clustering evaluation measure that assesses cluster definition through variance ratio between between-cluster variance and within-cluster variance.

• AKA: Calinski-Harabasz Index, Variance Ratio Criterion, CH Index.
• Context:
  • It can typically measure cluster separation using between-group dispersion.
  • It can typically evaluate cluster compactness through within-group dispersion.
  • It can typically compute F-statistic analogs for clustering validation.
  • It can often benchmark unsupervised legal document clustering tasks via quality scores.
  • It can often indicate optimal cluster numbers through score maximization.
  • It can range from being a Low Calinski-Harabasz Score Measure to being a High Calinski-Harabasz Score Measure, depending on its variance ratio value.
  • It can range from being a Spherical Calinski-Harabasz Score Measure to being an Elliptical Calinski-Harabasz Score Measure, depending on its cluster shape assumption.
  • It can range from being a Balanced Calinski-Harabasz Score Measure to being an Imbalanced Calinski-Harabasz Score Measure, depending on its cluster size distribution.
  • It can range from being a 2-Cluster Calinski-Harabasz Score Measure to being a Many-Cluster Calinski-Harabasz Score Measure, depending on its cluster count.
  • ...
• Examples:
  • Legal Clustering Benchmarks, such as:
    • ACORD Dataset CH Score showing high values.
    • CUAD Document CH Analysis for partition evaluation.
  • Algorithm Performance Comparisons, such as:
    • Hybrid Clustering CH Score exceeding baseline methods.
    • Topic Model CH Evaluation for quality assessment.
  • ...
• Counter-Examples:
  • Normalized Mutual Information Measure, which is an external measure.
  • Silhouette Coefficient Measure, which uses different calculation method.
  • Perplexity Measure, which evaluates language models.
• See: Internal Clustering Evaluation Measure, Clustering Evaluation Measure, Silhouette Coefficient Measure, Davies-Bouldin Index, Variance Analysis, F-Statistic, Cluster Compactness, Cluster Separation, ANOVA.

---