Skip to main content

Table 5 Outcome evaluation metrics and their meanings

From: Statistical power for cluster analysis

Value

Silhouette score meaning (computed from data)

Adjusted Rand index meaning (computed from data and ground truth)

< 0

Observations are closer to the centroid of another cluster than to the centroid of the cluster to which they were assigned

Below chance performance: random guessing would have provided more accurate labelling

0

Observations lie exactly between the centroids of their assigned cluster and the nearest of the other clusters

The number of labels assigned by the algorithm that overlap with the ground truth are at chance level

0.5–0.7

Evidence for clustering

 

0.7–1.0

Strong evidence for clustering

 

1

Perfect alignment of observations and their assigned cluster centroid

Perfect overlap between assigned clusters and ground truth