Skip to main content
. Author manuscript; available in PMC: 2022 Jun 1.
Published in final edited form as: J Biomed Inform. 2021 Apr 20;118:103788. doi: 10.1016/j.jbi.2021.103788

Fig. 2. Violin plots of adjusted rand index and silhouette width for simulated balanced and unbalanced binary data mixtures.

Fig. 2.

Varying data mixtures were clustered with 3 algorithms (hierarchical clustering “HC”, Partitioning Around Medoids “PAM”, and self-organizing maps “SOM”), 2 single-distance metrics (Manhattan and Euclidean distance) and 3 mixed-distance dissimilarity metrics (DAISY, Mercator, and Supersom). For both balanced (A) and unbalanced binary (C) simulations, all algorithms tested produce solutions with a range of ARI between 0 and 1, with improved performance with DAISY and Mercator with HC and SOM with the Manhattan distance. Among balanced data, DAISY and Mercator perform similarly. Among unbalanced binary data, DAISY with HC outperforms all other metrics and algorithms. Supersom produced superior SW (D) but low mean ARI compared to other methods.