Cluster-based dual evolution for multivariate systems

James, Nick; Menzies, Max

This paper proposes a cluster-based method to analyse multivariate systems that change over time. We apply this method to analyse the evolution of COVID-19 cases and deaths, partitioning data points into an appropriate number of clusters on each day to track both the total number of clusters and their changing constituency over time. This method can be used to track the trajectory of both the entire system as well as individual countries relative to the system. Applying our analysis to cases and deaths independently reveals a close relationship between the evolution of these two systems. With this in mind, we also develop a method to analyse the similarity and anomalies between two related multivariate systems in conjunction, allowing us to identify anomalous countries in the progression of cases to deaths.