Covariate-Informed Model-Based Clustering: An Application to Transportation Networks

13.03.2026 11:15 – 12:15

RESEARCH INSTITUTE FOR STATISTICS AND INFORMATION SCIENCE SEMINARS

ABSTRACT

Model-based clustering represents one of the primary application areas of Bayesian nonparametric methods, with mixture models serving as the prototypical framework for probabilistic clustering. These models naturally induce a prior distribution over partitions of the data, providing a flexible and principled approach to uncertainty quantification in clustering.
In this talk, I will briefly discuss the connection between mixture models and clustering, highlighting the role of the exchangeable probability partition function (EPPF) as the prior on the partition structure induced by the mixture specification.
In the second part, I will show how the EPPF can be modified to incorporate covariate information directly into the prior distribution on the clustering structure, so that observations with similar covariates have a higher prior probability of being clustered together. This leads to partition models in which similarity in covariate space increases the prior probability of co-clustering, while preserving the flexibility of the Bayesian nonparametric framework.
Motivated by real-world data on monthly subscriptions to the public transportation system of the Bergamo province (Italy), I will illustrate how properly transformed spatial covariates can be incorporated into a state-of-the-art stochastic block model, while allowing the contribution of covariates to be explicitly weighted.

Lieu

Bâtiment: Uni Mail

Boulevard du Pont-d'Arve 40
1205 Geneva

Room M 4220, 4th floor

entrée libre