Modeling and prediction of airborne infection risk in offices: A comprehensive CFD study and machine learning surrogate approach

Airborne transmission of respiratory viruses in office environments remains a critical public health concern. This study aims to quantify how office layout, A/C flow rate, and ventilation influence cough droplet dispersion and infection risk, and to develop a geometry-native surrogate model for rapid risk prediction. Using the supercomputer Fugaku, we conducted 896 simulations combining 32 diverse office layouts with 28 air conditioning and ventilation conditions. A Risk of Infection (RoI) index was defined based on the time-averaged quantity of cough-generated droplets suspended in the inhalation zone. The results revealed that seating arrangements significantly influence exposure risk; wall-facing seats generally exhibited lower and more stable RoI, whereas seats directly under air conditioning outlets showed high volatility depending on airflow settings. To enable rapid risk assessment without computationally expensive CFD, we developed a surrogate model using machine learning. Among ten regression algorithms compared, PointNet, which directly learns from 3D layout geometry, achieved the highest accuracy for the specific configurations investigated in this study. While the superiority of PointNet is demonstrated for this office-based dataset, its broader generalizability remains a target for future multi-facility verification.