Similarity maps and pairwise predictions for transmission dynamics of COVID-19 with neural networks

Abstract

On March 11, 2020, the World Health Organization declared COVID-19 as a pandemic. Since then, many countries have experienced the rapid transmission of this respiratory disease among their populations and have exercised many strategies to mitigate the spread of this disease. The prediction of the transmission dynamics serves important roles in designing mitigation strategies. However, due to the unknown characteristics of this disease, as well as the geographical and political factors, building efficient models of the dynamics for many countries is difficult. The objective of this study is to develop a transmission dynamics predictor that takes advantage of the time differences among many countries with respect to transmission of this disease, in that some countries experienced earlier outbreaks than others. The primary novelty of the proposed method is that, unlike many existing transmission predictors that require parameters based on prior knowledge of the epidemiology of past viruses, the proposed method only requires the transmission similarities between countries in the publicly available data for this current disease. In this paper, the viability and limitations of the proposed method are reported and discussed.

Keywords

• COVID-19
• Transmission dynamics
• Neural network
• Long short-term memory
• Time series prediction
• Topological representations