On Agglomerative Hierarchical Percentile Clustering
Fabrizio Durante, Aurora Gatto, Susanne Saminger‐Platz · Atlantis studies in uncertainty modelling/Atlantis Studies in Uncertainty Modelling · 2021
Cluster analysis aims at grouping objects represented by some feature vectors and as such revealing insight into subset structures among the considered objects.However, in many cases, the observations are subject to experimental errors and/or uncertainty.In such a case, a popular way is to summarize first the information about each object and, then, aggregate the objects via some cluster algorithm.The percentile clustering by Janowitz and Schweizer, instead, considers the whole distribution of observed features and, only afterwards, aggregates them.Here, we revisit this approach in an agglomerative clustering perspective.Moreover, we perform a simulation study showing some finite sample performance of the algorithm.Some case studies illustrate the advantages of the whole methodology.