Classification of a COVID-19 dataset by using labels created from clustering algorithms

Abstract

Novel coronavirus (COVID-19) is a newly discovered infectious disease that has received much attention in the literature because of its rapid spread and daily global deaths attributable to such disease. The White House, together with a coalition of leading research groups, has published the freely available COVID-19 Open Research Dataset to help the global research community apply the recent advances in natural language processing and other AI techniques in generating novel insights that can support the ongoing fight against this disease. In this paper, the hierarchical and k-means clustering techniques are used to create a tool for identifying similar articles on COVID-19 and filtering them based on their titles. These articles are classified by applying three data mining techniques, namely, random forest (RF), decision tree (DT) and bagging. By using this tool, specialists can limit the number of articles they need to study and pre-process these articles via data framing, tokenisation, normalisation and term frequency-inverse document frequency. Given its 2D nature, the dimensionality of this dataset is reduced by applying t-SNE. The aforementioned data mining techniques are then cross validated to test the accuracy, precision and recall performance of the proposed tool. Results show that the proposed tool effectively extracts the keywords for each cluster, with RF, DT and bagging achieving optimal accuracies of 98.267%, 97.633% and 97.833%, respectively.