Health and environmental data are often heterogeneous, multivariate, longitudinal, temporally correlated, and sometimes incomplete. L2S develops statistical learning methods to integrate these structures, reveal associations, and build interpretable models.
This research is conducted with medical and environmental partners. It addresses multimodal cohorts, health data warehouses, omics data, and time series arising from pollution monitoring and epidemiology.
We develop multiblock and multiway methods to jointly analyze several data sources describing the same individuals, including clinical, biological, imaging, and sensor data. Extensions of RGCCA to tensor data preserve their natural organization and identify components shared by, or specific to, the different blocks.
Medical data are frequently collected at irregular times, with few measurements per patient. We propose functional and longitudinal methods that can exploit large cohorts despite this sparse observation pattern. Our work also investigates structural equation models to move beyond correlation and study relationships between observed and latent variables.
We study time-series models for count data, multivariate observations, and outliers. These methods are used to analyze relationships between air pollution and health, model respiratory diseases, and characterize the effects of environmental exposure over time.
L2S develops the open-source RGCCA software, a unified environment for multiblock data analysis. The package provides analysis methods, graphical displays, and tools to assess the robustness and significance of results. It is also distributed through CRAN.
Professor – CentraleSupélec
Signaux et statistiques – Signal & Stat
arthur.tenenhaus@l2s.centralesupelec.fr
Bât. Breguet .
