A Dual Index Machine Learning Framework Integrating Structural Vulnerability and Human Mobility for Dynamic Pandemic Risk Assessment

The COVID-19 pandemic has highlighted the urgent need for spatially and temporally precise tools to assess community vulnerability and inform equitable public health responses. This study proposes a dual-index framework that integrates structural and mobility risk factors to support dynamic pandemic modeling. We first construct the Pandemic Responsive Vulnerability Index (PRVI) using 53 sociodemographic variables from the American Community Survey (ACS), weighted by their empirical correlation with COVID-19 case rates in San Diego County. Building on this static foundation, we develop the Dynamic Pandemic-Responsive Index of Spatio-temporal Mobility and Vulnerability (D-PRISM) by incorporating lagged mobility metrics derived from SafeGraph data into a Random Forest predictive model. The model can accurately capture spatial patterns and localized outbreaks across dynamic temporal windows that static index alone fails to detect. The PRVI and D-PRISM serve complementary functions: PRVI profiles long-term structural vulnerability, while D-PRISM enables real-time risk detection. Together, they offer a scalable and transferable approach to modeling pandemic risk. The framework can be applied beyond COVID-19 to other infectious diseases or pandemics that show spatio-temporal dynamics. By using divergent high-resolution data, machine learning, this work advances the field of precision public health and provides insights for targeted interventions.