Divisive Property-Based and Fuzzy Clustering for Sequence Analysis

Abstract

This paper discusses the usefulness of divisive property-based and fuzzy clustering for sequence analysis. Divisive property-based clustering provides well-defined clustering membership rules. Aside from significantly simplifying interpretations of clustering, it is also useful when one plans to use the same typology in other samples or studies. We further enrich the methods by proposing new sets of sequence features that can be automatically extracted and used in the procedure. We then discuss the use of fuzzy clustering, where sequences belong to each cluster with an estimated membership strength. This approach is particularly useful when some sequences are thought to lie between two (or more) sequence types (i.e., hybrid-type sequences) or when only a weak structure is found in the data. This paper also discusses several methods by which to visualize a fuzzy clustering solution, and analyzes them with regression-like approaches. It also introduces R code to run all the discussed analyses; all the proposed developments are made available in the WeightedCluster R package.