DefinePK

Abstract

Link prediction is an important area of research in complex networks, helping to understand the numerous relationships between nodes based on their topological structures. Recently, link prediction has been extensively used to study the spread of COVID-19. Given the diffusion of the virus from one location to another, it is vital to classify affected locations and predict the links related to the virus's spread. We propose a novel approach for the COVID-19-infected locations based on complex network theory to analyze and explore link prediction. For this, we construct a weighted two-mode network from the COVID-19 dataset, comprising confirmed case records of affected locations during specific weeks. The network comprises nodes representing locations and weeks, these two nodes are linked if a COVID-19 case is confirmed within a location during a particular week. The weight of the links corresponds to the frequency of COVID-19 cases. We apply five local and four global link prediction methods and evaluate their results using the ROC curve. We select the most appropriate link prediction method for the observed network based on our evaluation. The experimental results show that the ACT method has outperformed the other eight local and global similarity methods for the COVID-19 location network. Consequently, our findings reveal better accuracy, signifying the effectiveness and applicability of the ACT link prediction method.