[MS07] From Theory to Data-Driven Applications: Understanding Infectious Diseases in a Lower-Middle Income Country Using Compartmental Modelling

[MS07-05] Can Transmission Rates Vary? Using Physics-Informed Neural Networks (PINN) for Modeling COVID-19 Epidemiological Dynamics

Aldie Alejandro¹, *Jeric Briones¹, Maria Regina Justina Estuar¹, Elvira de Lara-Tuprio¹, Timothy Robin Teng¹, Mark Anthony Tolentino¹ (1. Ateneo de Manila University (Philippines))

Keywords:

compartmental models,COVID-19,physics-informed neural networks,time-dependent parameters,transmission dynamics

While compartmental models have been widely used to model the transmission of different infectious diseases at population level, such as COVID-19, these models usually have the limiting assumption that flow parameters are constant. To relax such assumptions, machine learning methods have been used to explore and understand further the dynamics in these compartmental models. For example, novel application of physical models in physics-informed neural network (PINN) COVID-19 is deemed effective in simultaneously modeling the complexity of transmission dynamics while inferring time-varying parameters and estimating unobserved compartments using only limited data. In this work, we developed a PINN architecture that models the COVID-19 transmission in a Philippine city. Using daily case data from 2022 to 2024, deep neural networks were incorporated to model time-varying parameters and compartmental values, thereby allowing for some pertinent parameters such as the transmission rate, detection rate, and recovery rate to be time-varying. Results have shown that the model was able to accurately forecast 30-day daily active cases in the city. The model also has the potential to provide insights regarding the transmission dynamics in COVID-19 data such as estimating unobserved compartmental flow and forecasting transmission peaks. These insights can then be used for better disease surveillance and pandemic preparedness.