ClusterAccuracy: Clustering Accuracy

In FCPS: Fundamental Clustering Problems Suite

Clustering Accuracy

Description

Clustering Accuracy

Usage

ClusterAccuracy(PriorCls,CurrentCls,Fast=TRUE,K=9)

Arguments

PriorCls	Ground truth,[1:n] numerical vector with n numbers defining the classification. It has k unique numbers representing the arbitrary labels of the clustering.
CurrentCls	Main output of the clustering, [1:n] numerical vector with n numbers defining the classification. It has k unique numbers representing the arbitrary labels of the clustering.
Fast	If TRUE, uses ClusterAlignLabels, for which clue solves a one-to-one assignment problem. If FALSE, searches over label permutations.
K	Used only when Fast = FALSE: the maximum number of classes used in the permutation computation.

Details

Here, accuracy is the proportion of observations whose labels agree after the candidate labels have been optimally aligned with the reference labels. Because clustering algorithms assign labels arbitrarily, the alignment that gives the highest accuracy is used [Thrun et al., 2018]. In contrast to ClusterMCC, the original labels may therefore be arbitrary. Accuracy is most informative when clusters are balanced or nearly equal in size; otherwise, consider ClusterMCC.

In contrast to the F-measure, "Accuracy tends to be naturally unbiased, because it can be expressed in terms of a binomial distribution: A success in the underlying Bernoulli trial would be defined as sampling an example for which a classifier under consideration makes the right prediction. By definition, the success probability is identical to the accuracy of the classifier. The i.i.d. assumption implies that each example of the test set is sampled independently, so the expected fraction of correctly classified samples is identical to the probability of seeing a success above. Averaging over multiple folds is identical to increasing the number of repetitions of the Binomial trial. This does not affect the posterior distribution of accuracy if the test sets are of equal size, or if we weight each estimate by the size of each test set." [Forman/Scholz, 2010]

Value

A numeric scalar between zero and one.

Author(s)

Michael Thrun

References

[Thrun et al., 2018] Michael C. Thrun, Felix Pape, Alfred Ultsch: Benchmarking Cluster Analysis Methods in the Case of Distance and Density-based Structures Defined by a Prior Classification Using PDE-Optimized Violin Plots, ECDA, Potsdam, 2018

[Forman/Scholz, 2010] Forman, G., and Scholz, M.: Apples-to-apples in cross-validation studies: pitfalls in classifier performance measurement, ACM SIGKDD Explorations Newsletter, Vol. 12(1), pp. 49-57. 2010.

See Also

ClusterMCC, ClusterAlignLabels

Examples

#Influence of random sets/ random starts on k-means

data('Hepta')
Cls=kmeansClustering(Hepta$Data,7,Type = "Hartigan",nstart=1)
table(Cls$Cls,Hepta$Cls)
ClusterAccuracy(Hepta$Cls,Cls$Cls)



data('Hepta')
Cls=kmeansClustering(Hepta$Data,7,Type = "Hartigan",nstart=100)
table(Cls$Cls,Hepta$Cls)
ClusterAccuracy(Hepta$Cls,Cls$Cls)

FCPS documentation built on Oct. 3, 2026, 9:06 a.m.