Sarcouncil Journal of Applied Sciences Aims & Scope

Abstract: The COVID-19 pandemic highlighted the need for models that are both mechanistic and data-driven. In this study, we develop and analyze a hybrid machine learning–ODE framework for predicting the spread of COVID-19 variants in the United States and for assessing public health interventions. We formulate a susceptible–exposed–infectious–recovered (SEIR) model with a time-varying transmission rate β(t) that is learned from covariates using a recurrent neural network. The covariates include mobility indices, vaccination coverage, and variant prevalence. We present basic qualitative properties of the model, derive the effective reproduction number, and describe the parameter estimation strategy. The hybrid framework reproduces the timing and relative magnitude of major epidemic waves and improves short-term forecast accuracy during periods of structural change, such as the emergence of the Delta and Omicron variants. We further show how the model can be used to simulate alternative intervention scenarios, such as increased booster coverage or changes in mobility. The proposed framework provides a concrete example of how hybrid ML–ODE models can support data-driven public health planning in the U.S.