A0912
Title: Semi-supervised time series clustering with copulas
Authors: Fabrizio Durante - University of Salento (Italy) [presenting]
Alessia Benevento - University of Salento (Italy)
Roberta Pappada - University of Trieste (Italy)
Abstract: Clustering algorithms for time series data play a crucial role as a pre-processing step in building multivariate stochastic models. The focus is on copula-based clustering methods that effectively capture comovements among different time series, independently of their marginal distributions. It begins by reviewing several approaches to clustering random variables using various dissimilarity indices derived from association measures such as Kendall's tau, Spearman's rho, and tail dependence coefficients. Novel algorithms are then introduced that enhance the clustering process by integrating additional deterministic information about the variables within a semi-supervised learning framework. These methods are particularly well-suited for geo-referenced time series, where incorporating spatial context into the dissimilarity measure is essential to uncover meaningful patterns in the data.
