Abstract
During the COVID-19 pandemic, every government faced the same impossible optimization problem: how to reduce viral transmission without destroying the economy. Closures, gathering restrictions, transportation limits, and healthcare mandates all reduce case counts — but each comes with economic costs, and the most stringent measures are often the most expensive. Simply applying maximum restrictions everywhere is unrealistic. Decision-makers need to know the optimal stringency level for each policy type, given their specific budget and resource constraints.
This is a multi-objective optimization problem, and we treated it as one. We developed a computational framework that combines Gated Recurrent Units (GRU) for case forecasting with an evolutionary-based meta-heuristic method for intervention prescription. The GRU model predicts the number of new cases given a set of intervention policies at specified stringency levels, and the evolutionary algorithm searches the policy space for configurations that are Pareto-optimal — meaning that no policy mix can reduce cases without increasing cost, or reduce cost without increasing cases. The result is not a single recommendation but a set of trade-off solutions, allowing decision-makers to choose the balance that fits their context.
This framework was published in Neural Computing and Applications (2022) and demonstrated the ability to identify policy configurations that outperform single-objective approaches. Our team was also selected as a finalist in the XPRIZE Pandemic Response Challenge, a global competition for pandemic intervention strategies.
Additional studies in this line analyzed the impact of various intervention policies across European countries, incorporating weather data and policy stringency levels. A related project, mentored with an undergraduate student, resulted in a conference paper at the IISE Annual Conference and earned the third-best undergraduate poster award at the IMSE Research Symposium.