This study provides a practical guide to using the Davies-Bouldin Index (DBI), a popular metric for internally evaluating clustering results. While previous studies applied DBI across results on various datasets and generated by several algorithms, no comprehensive guidelines exist for effective use of this coefficient. We systematically analyze DBI’s behavior through experiments on both synthetic and real-world medical datasets, including results obtained through k-means, DBSCAN, and hierarchical clustering. Our results highlight DBI’s strengths and limitations, such as sensitivity to cluster imbalance, non-compact clusters, and high-dimensional data. Based on these findings, we propose actionable recommendations and best practices for unsupervised machine learning researchers, including optimal algorithm choices and parameter settings. This study bridges the gap between theoretical knowledge and practical application, offering a clear manual of use for the Davies-Bouldin Index in diverse clustering tasks.
Anfossi et al. (Fri,) studied this question.