Table 3.

Measures of clustering quality. The measures are assigned into three different groups based on principal component analysis (see Results section).

Measure	Brief description	Range	Group
Adjusted mutual information (AMI)	A variation of mutual information between two clustering partitions, adjusted for the effect of chance agreements between partitions	0 to 1	1
Adjusted Rand index (ARI)	A variation of Rand index as a measure of the percentage of correct matches, adjusted for the effect of chance agreements between partitions	0 to 1	1
F-measure	A measure of accuracy that balances both the precision and recall	0 to 1	1
Variation of information (VI)	An information-based measure that behaves like a true distance, with zero representing equality of the two partitions	0 to infinity	1
Homogeneity	Entropy-based measure that quantifies whether only those data points that are members of the same class are assigned to the same cluster.	0 to 1	2
Majority	Proportion of the data in the largest cluster	0 to 1	2
Silhouette	Clustering fitness that measures whether each data point belongs unambiguously to the cluster to which it has been assigned	−1 to 1	3