Similarity Pattern Discovery for COVID-19 Using Sequential Decision Tree

Li, Z.; Liang, F.

Decision tree algorithm that handles single classes is often used in biology and biomedicine. However, biological datasets frequently involve multiple classes related by a hierarchy, and each class can contain multiple labels, which cannot be predicted by the commonly used decision tree algorithm as classes used to define one higher hierarchical level would not appear again in the lower one. In this paper, a novel sequential decision tree (SDT) algorithm for hierarchical multi-label classification is proposed, and the SDT application to COVID-19 is examined. Comparing to the existing algorithms, SDT has good interpretability to be used on the population-demographic-COVID-19 (PDC) dataset. The goal of research is to compute the similarity of countries under the influence of COVID-19 using PDC as an input dataset. The hypothesis is made that the most similar countries predicted from the PDC dataset, which is defined by the DT leaves branched from the same end node, also have similar COVID-19 evolutions. It is defined by the small difference in infected, recovered, and dead time series curves of the countries. From data and statistical analysis, the hypothesis is supported. © 2021 ACM.