Cluster evaluation metrics in ML Python - Cheat Sheet & Quick Revision

Recall & Review

beginner

What is the purpose of cluster evaluation metrics?

Cluster evaluation metrics help us measure how well a clustering algorithm groups data points. They tell us if clusters are tight and well-separated.

intermediate

Explain the Silhouette Score in clustering.

The Silhouette Score measures how similar a point is to its own cluster compared to other clusters. Scores near +1 mean good clustering, near 0 means overlapping clusters, and negative means wrong cluster assignment.

intermediate

What does the Davies-Bouldin Index indicate?

The Davies-Bouldin Index measures average similarity between clusters. Lower values mean clusters are compact and far apart, which is better.

intermediate

Describe the difference between internal and external cluster evaluation metrics.

Internal metrics use only the data and cluster labels to evaluate quality (e.g., Silhouette Score). External metrics compare clusters to known true labels (e.g., Adjusted Rand Index).

intermediate

What is the Adjusted Rand Index (ARI) used for?

ARI measures similarity between the clustering result and true labels, adjusting for chance. It ranges from -1 to 1, where 1 means perfect match.

Which cluster evaluation metric ranges from -1 to 1 and measures similarity to true labels?

A Silhouette Score

B Adjusted Rand Index

C Davies-Bouldin Index

D Calinski-Harabasz Index

The Adjusted Rand Index compares clustering results to true labels and ranges from -1 to 1.

A high Silhouette Score indicates:

A Clusters overlap heavily

B Clusters are poorly formed

C Clusters are well separated and compact

D Clusters have many outliers

High Silhouette Scores mean points are closer to their own cluster than others, showing good clustering.

Which metric should be minimized for better clustering?

A Davies-Bouldin Index

B Adjusted Rand Index

C Silhouette Score

D Homogeneity Score

Lower Davies-Bouldin Index values indicate better cluster separation and compactness.

Internal cluster evaluation metrics require:

A Human expert input

B True labels of data

C External validation data

D Only data and cluster assignments

Internal metrics evaluate clusters using only the data and the clusters found, without true labels.

Which metric compares clustering results to known labels adjusting for chance grouping?

A Adjusted Rand Index

B Calinski-Harabasz Index

C Davies-Bouldin Index

D Silhouette Score

Adjusted Rand Index adjusts for chance and compares clustering to true labels.

Explain how the Silhouette Score helps evaluate clustering quality.

Think about how close points are to their own cluster versus others.

Describe the difference between internal and external cluster evaluation metrics with examples.

Consider whether true labels are needed.