Database Reference
In-Depth Information
External evaluation metrics
Since clustering can be thought of as unsupervised classification, if we have some form of
labeled (or partially labeled) data available, we could use these labels to evaluate a cluster-
ing model. We can make predictions of clusters (that is, the class labels) using the model
and evaluate the predictions against the true labels using metrics similar to some that we
saw for classification evaluation (that is, based on true positive and negative and false pos-
itive and negative rates).
These include the Rand measure, F-measure, Jaccard index, and others.
Note
See
http://en.wikipedia.org/wiki/Cluster_analysis#External_evaluation
for more informa-
tion on external evaluation for clustering.
