A Study on Unified Role of Warning Signs and Signal-SIR Model Using DNN to Predict Epidemic Transmission

Abstract

This study extends the classical Susceptible-Infected-Recovered (SIR) model by integrating adaptive behaviors and policy interventions during epidemics through the Signal-SIR model. Here the susceptible population is divided into two groups: individuals who adhere to health regulations (AD strategy) and those who do not (NAD strategy). The model simulates the dynamic interaction between government signals and public behavior where it utilizes replicator dynamics to explore how health warnings influence population responses. It also introduces chicken game payoffs to analyze the redistribution of risks between compliant and non-compliant individuals. To optimize model parameters and explain time-varying dynamics, deep neural networks (DNNs) has been employed alongside Stochastic Gradient Descent. We establish a loss function that quantifies the discrepancies between observed data and model predictions. Simulation results indicate that enhanced adaptive behavior, driven by enhanced adherence to health regulations, significantly reduces the spread of infection. Therefore, it leads to lower infection peaks and higher recovery rates. This paper highlights the critical role of adaptive strategies in public health policy and provides a data-driven framework for effectively forecasting and managing epidemic dynamics. 

J. Bangladesh Math. Soc. 45.1 (2025) 01–15