Efficient Computation and Visualization of Multiple Density-Based Clustering Hierarchies

HDBSCAN*, a state-of-the-art density-based hierarchical clustering method, produces organization of clusters in dataset w.r.t. parameter mpts. While small change mpts typically leads to the structure, choosing “good” value can be challenging: depending on data distribution, high or low may more appropriate, and certain reveal themselves at different values. To explore results for range values, one has run HDBSCAN* each independently, which computationally impractical. In this paper, we propose an approach efficiently compute all hierarchies values by building upon from computational geometry replace HDBSCAN*'s complete graph with smaller equivalent graph. An experimental evaluation shows that our obtain over hundred cost running about twice, corresponds speedup than 60 times, compared independently many times. We also series visualizations allow users analyze collection along case studies illustrate how these analyses are performed.