Network Inference-Based Prediction of Epidemics: A Case Study on Mexican State COVID-19 Infection Counts

With the world recently having suffered from the global COVID-19 pandemic, it created a necessity to predict the spread of this virus and that of possible future epidemics. Predicting the spread of the virus, and understanding the way a virus interacts within individuals of a population, can contribute to its understanding as well as the effectiveness of counter measures. In an attempt to make these predictions it is possible to use the existing COVID-19 data, including time series of infection counts gathered during the pandemic. A method to infer an interaction network from this data, and make predictions on the future dynamics of this network, is the Network Inference-based Prediction Algorithm (NIPA). This paper aims to infer the COVID-19 interaction network from the daily infection data of the states of Mexico using NIPA. The SIR (Susceptible Infected Removed) epidemic model is applied to capture the dynamics of the COVID-19 spread within each state. We exploit the inferred interaction network in an attempt to estimate the interaction patterns between states, and compare those with the observations from past COVID-19 outbreaks. Finally, we assess the results produced by the inferred infection matrix, and explain how they reflect on different aspects of a virus spreading in the real world, such as via international visitors and tourism.