The Analysis of Hybrid Machine Learning Approach for Evaluating the Non-Pharmaceutical Intervention on Transmission of COVID-19 in China

Abstract

COVID-19 is a severe disaster for human society. There are several models to predict the transmission of COVID-19. This study assessed a hybrid modelling framework that integrates a time-varying SEIRD model with a LSTM neural network to evaluate the effects of non-pharmaceutical interventions (NPIs) on COVID-19 transmission in China. The model is trained on daily provincial-level data from 2020 to 2023, including confirmed cases, policy indicators, and mobility indices. Through the LSTM, the framework captures both mechanistic epidemic dynamics and behavioral responses to interventions. Compared to baseline models, the hybrid approach yields lower RMSE and MAPE across 31 provinces and regions, particularly during periods of policy shifts and regional outbreaks. This approach provides a flexible and interpretable tool for forecasting regional outbreaks and assessing the effectiveness of NPIs, supporting data-driven public health decision, which is meaningful for preventing the further spread of the virus and provide the experience for dealing with similar epidemics.