Socioeconomic inequality as a determinant of divergent human behavioral responses to epidemic evolution

Understanding how collective behaviors evolve during outbreaks is essential for accurate prediction and effective intervention design, yet it remains insufficiently explored across diverse socioeconomic contexts. We propose a Behavior-Coupled Physics-Informed Neural Network (BC-PINN), establishing a flexible, data-driven framework to learn heterogeneous behavioral responses across varying socioeconomic environments. Following BC-PINN training, we further employ symbolic regression to recover mechanistic functions governing behavioral evolution, enabling interpretable learning and pattern identification. Applying our framework to COVID-19 data across 58 countries, we uncover two distinct behavioral evolution patterns strongly associated with levels of socioeconomic development. Specifically, we observe a significant transition around a Human Development Index (HDI) of 0.8, where countries above this value exhibit Benefit-Driven Behavior (BDB), while those below display Barrier-Inhibition Behavior (BIB). Simulation results suggest that if countries exhibiting BIB transitioned to BDB, both epidemic peaks and cumulative infections at peak are consistently decreased in several countries. Moreover, within the BDB group, behavioral response intensity positively correlates with GDP per capita. These findings underscore the profound impact of economic inequality on collective behavioral responses during pandemics, offering valuable insights and practical guidance for crafting tailored public health strategies across diverse socioeconomic contexts.