Applied & Computational Mathematics seminar: New data-driven machine learning insights into vaccination scare behavior

Speaker:  Igor Erovenko, UNC Greensboro

Title:  New data-driven machine learning insights into vaccination scare behavior

Abstract:   Vaccine scares, such as the 1970s pertussis crisis in England and Wales, severely undermine infectious disease control. While evolutionary game theory provides a robust framework for modeling vaccination behavior, calibration is often hindered by unobservable parameters. We address this challenge by applying physics-informed Kolmogorov-Arnold Networks (PIKANs) to a coupled behavioral-epidemiological model of pertussis. By embedding the governing differential equations into the network’s loss function, we ensure theoretical consistency while simultaneously estimating system states and parameters from historical vaccination and incidence data. Notably, the KAN architecture allowed us to discover the functional form of perceived vaccine risk without a priori structural assumptions, yielding an exceptional model fit to vaccination data. This discovered function revealed localized fluctuations, which are undetectable by standard approaches, that align precisely with historical periods of intensified negative vaccine media coverage. Our results demonstrate that PIKANs are a powerful tool for uncovering latent behavioral drivers, offering a new pathway for evidence-based public health policy.

Time:  Friday, April 3, 1:30pm – 2:30pm
Place:  Exploratory Hall, room 4106