Predicting COVID-19 Transmission in Southern California with Machine Learning Methods

The COVID-19 pandemic has posed significant chal-lenges to global health, economies, and social structures since its emergence in late 2019. Southern California, a region with unique characteristics and diverse communities, has experienced difficul-ties in controlling the virus's spread. Understanding the factors that influence COVID-19 transmission in Southern California is crucial for informing public health policies and mitigating the virus's impact on the communities. In this study, we collected weekly zip code level data from March to December 2020 in Southern California and applied various machine learning algorithms, such as graph neural networks (GNN), multi-layer perceptrons (MLP), and gradient boosting trees (XGBoost), to model the spread of COVID-19. Our main contributions include the development of various COVID-19 forecasting models that incorporate socioeconomically significant variables and zip code level mobility data, coupled with variable impact assessments to provide valuable insights for informing future public health policies and strategies. Our predictive model not only attains high levels of accuracy in forecasting but also facilitates the understanding of how each variable contributes to the final outcome. These findings have the potential to enhance our understanding of COVID-19 transmission patterns and inform targeted mitigation strategies in the region.