Calinski-Harabasz Index

INFO

Measures the ratio of the sum of between-cluster dispersion to within-cluster dispersion. A higher index value indicates better clustering.

How It Works

$$CH=\frac{Tr(B_k)}{Tr(W_k)}\cdot \frac{N-k}{k-1}$$

• $Tr(B_k)$: Between-cluster dispersion
• $Tr(W_k)$: Within-cluster dispersion
• $N$: Total number of samples
• $k$: Number of clusters

What to Look For

• Higher CH = better-defined clusters
• Sensitive to cluster density and separation
• Works well for convex clusters

Application Models

• K-Means Clustering
• Hierarchical Clustering
• Density-Based Spatial Clustering of Application with Noise (DBSCAN)
• Gaussian Mixture Models