Abstract:
In this talk I'll present a novel approach for clustering called CUBT (Clustering Using Binary Trees). It is inspired from Breiman's Classification and regression trees adapted for unsupervised context. CUBT, initially designed for continuous variables, has been adapted for nominal variables, and recently for mixed type data. Its most interesting feature is its interpretability and the possibility of ranking variables significant for the clustering. Actual work aims its extension to compositional data.