B1360
Title: Mixtures of product partition models with covariates to cluster blood donors
Authors: Raffaele Argiento - Università degli Studi di Bergamo (Italy) [presenting]
Alessandra Guglielmi - Politecnico di Milano (Italy)
Riccardo Corradin - University of Nottingham (United Kingdom)
Abstract: The challenge of accurately predicting the time gaps between successive blood donations is addressed. We propose a novel class of Bayesian nonparametric models for clustering to achieve this. These models can predict new occurrences while considering relevant information about the individuals in the sample, such as their personal characteristics. To achieve this, a prior is introduced that facilitates the random partitioning of the sample individuals. The prior encourages grouping two individuals into the same cluster if they share similar covariate values. By doing so, the class of product partition models is extended with covariates (PPMx) by considering a mixture of PPMx with similarity functions reflecting the density of a cluster. It is demonstrated that incorporating covariate information in the prior specification leads to improved performance in predicting future events and facilitates the interpretation of estimated clusters with respect to the covariates in the context of blood donation applications.
