Copula-based clustering algorithms
Clustering algorithms aim at identifying association and dependence within a vector of n ≥ 3 random variables and creating sub-vectors of variables that are considered to be similar according to a given criterion. Such methods are particularly useful as a pre-processing step in the building of a multivariate stochastic model and can be interpreted as an unsupervised learning method. In this talk, we first review various methodologies to cluster random variables according to different notions of similarity, as derived from measures of association like Kendall’s tau, Spearman’s rho, and tail dependence coefficients. Then, we present some novel algorithms that use some additional information about the involved random variables in order to guide the clustering process in a semi-supervised learning setting [1, 2]. Such information can be particularly related to geographic information about the sites where the variables are observed and, hence, is particularly useful for geo-referenced data.
Area: IS19 - Dependence Modeling (Elisa Perrone)
Keywords: Copula, clustering algorithm
Please Login in order to download this file