Publications:
If your are using any of the material below please cite the corresponding publication:
- (pending)
Description:
GOI provides a set of indices for absolute cluster validation and for the interpretation of the dataset context based on geometrical properties of the multidimensional data. In addition to cluster masses, inter- and intra-cluster distances, GOI comprises global (G) and individual (oi) overlap indices, density evaluations, cluster multimodality detection and cluster kinship recognition.
Software, scripts, tools:
GOI indices are implemented within the CVTbed v1.0 (Testbed for Cluster Validation) for the MATLAB environment. CVTbed v1.0 can be downloaded from the GOI GitHub repository. Some functions require the creation of synthetic datasets with MDCGen, which can be downloaded from the MDCGen GitHub repository.
Datasets, experiments:
The CVTbed implementation in the current GOI GitHub repository includes the experiments conducted in the paper:
(publication pending)
They are:
-
Oct 2016 - Comparison of clustering performances.
-
Oct 2016 - Calculation of G and other crisp and fuzzy validity indices (Silhouette, Partition coefficient, etc.).
-
Oct 2016 - Tests for GOI-density dependency on dimensionality.
-
Oct 2016 - Tests for multimodality detection within clusters based on kernel density estimations.
-
Oct 2016 - Tests for the comparison of cluster validation techniques with multidimensional datasets.
Example figures:
Images show the example of a dataset clustered with the optimal k=20. Time series show various indices for clustering solutions with different initial 'k', from k=5 to k=20. Last image contains the three GOI indices.