GEVA: geometric variability-based approaches for identifying patterns in data (Q2430241)

Four new general approaches to clustering based on geometric variability (GEVA) are presented. One algorithm is for partitioning a group into \(k\) subgroups, for some specified \(k>1\). The other three algorithms are agglomerative. GEVA-clustering algorithms require only a distance between individuals and can be applied to general studies involving binary, ordinal and countable data. Properties of GEVA clustering procedures are observed. The GEVA approaches are illustrated using hypothetical and real data on Drosophila chromosomal inversions. It is shown that for Drosophila subobscura the results are better than those obtained by standard hierarchical clustering.