Optimal control and computational tools in epidemiology: from health/economics dilemmas and human behaviour to machine learning applications

This thesis studies how optimal control theory can support pandemic preparedness by balancing epidemiological benefits against the social and economic costs of intervention. Using a calibrated COVID-19 compartmental model based on Italy’s first wave, it examines how the best social-distancing strategies change with three key factors: the weight given to indirect costs, population adherence to non-pharmaceutical interventions, and the timing of policy activation. Across several extensions of the model, including finite hospital capacity, continuous importation of infective travellers, and behavioural fatigue, the thesis shows that optimal policies can shift sharply from strong suppression to weak control when indirect costs are overvalued. It also highlights trade-offs between intensity and duration of measures, the role of hospital saturation, and the importance of behavioural feedbacks. The work combines open-loop optimal control, reinforcement learning, and interpretable forecasting to build a practical catalogue of preparedness policies and to test how robust they remain in more realistic, data-driven settings.