Disease transmission in dynamic social networks constructed by reinforcement learning-driven preventive game

Reinforcement learning delves into the intrinsic behaviors of multi-agents, while complex networks reveal their interactive relationships. In real life, an individual can be considered as a sophisticated multi-agent, whose behavioral choices are influenced by a myriad of factors. When contagious diseases penetrate a population, the adoption of preventive measures represents a manifestation of behavioral selection. This paper introduces reinforcement learning into the multi-agent decision within the network, leveraging the Q-learning algorithm to train individual decision-making. At each time step, individuals select the next preventive strategy based on their utility and strategies, and these choices subsequently alter the network structure, thereby influencing the prevalence of the disease. In addition, the incorporation of additional government incentives prompts individuals to opt for advanced preventive strategies, effectively containing disease transmission within the population. The construction of this model not only showcases the co-evolutionary process between network structure and disease transmission but also validates its efficacy through theoretical analysis and applications in real-world networks, offering a novel perspective towards the epidemic control. • Reinforcement learning is used to study how individual prevention choices affect disease spread in networks. • Effective government rewards and containment are designed to better prevent the epidemics. • Prevention strategies render networks more resilient and epidemic outbreaks more difficult.