Etudiant : ETTOUBAL ZAYED

Filière : Modélisation Stochastique et Statistique Décisionnelle (MSSD)

Encadrant : Pr. EL KHALIFI MOHAMED

Annèe : 2026

Résumé : The COVID-19 pandemic highlighted how difficult it is to control an epidemic while decisions must be made under uncertainty. This thesis presents an adaptive framework that combines stochastic epidemic modelling, sequential Bayesian inference and reinforcement learning to support public health interventions. An extended stochastic SEIR model, formulated as a hidden Markov model, is used to describe disease transmission, vaccination and intensive care dynamics, while sequential Monte Carlo continuously updates the epidemic state as new data become available. Based on these updated estimates, the control problem is formulated as a Markov decision process and solved using reinforcement learning. Two complementary decision strategies are investigated, one favouring interpretability and the other focusing on performance. The proposed framework is validated with real COVID-19 intensive care data, showing its ability to support reliable and data-driven epidemic control under uncertainty