dbcv

In order to validate density-based clustering results, relative CVIs, such as Silhouette and Dunn cannot be directly applied. Simply put, such indices are based on distances between objects, not taking density information into account. Density-Based Clustering Validation (DBCV) is a relative validity criterion that can be applied in such a scenario. DBCV can be employed to select the best configuration of parameters for DBSCAN, for instance. The measure is described here.

If you want to use the measure, Matlab source code for DBCV provided here.

The synthetic data sets used in the paper (shown below), are also available in the github repository.

Enjoy!