I run algorithms like HDBSCAN and MeanShift, and compare them with K-means and GaussianMixture. The first two algorithms do not return cluster labels for some elements in the dataset. My question is - how to correctly compare such different algorithms? Do I have to remove elements with no cluster labels from the dataset before performance metrics evaluation? My metrics are the Silhouette score, Davies-Bouldin score, and Calinsky-Harabasz Index.
Have a look at https://arxiv.org/pdf/1905.05667.pdf
Also you may follow this RG thread : https://www.researchgate.net/post/Evaluation_methods_for_a_clustering_techniques2
https://arxiv.org/pdf/1905.05667
http://vargas-solar.com/data-centric-smart-everything/hands-on/unsupervised-learning-comparing-clustering-methods/
Vladislav Bespalov
Please check below
Article Flexible Affinity Matrix Learning for Unsupervised and Semis...
- Badges
- Science topic