Effectiveness of deep recurrent neural networks in classifying COVID-19-related respiratory conditions using clinical reports

Abstract Countries all around the world have been severely affected by the novel coronavirus, popularly known as COVID-19, causing hundreds of thousands of unfortunate deaths without having found a widespread immunization method. It is extremely crucial for healthcare officials and the Governments of various countries to be able to differentiate the possible coronavirus strains to provide appropriate care and treatment. This paper proposes an effective way to set apart coronavirus strains into classes such as severe acute respiratory syndrome (SARS), acute respiratory distress syndrome (ARDS), and COVID-19 using machine learning algorithms and recurrent neural networks (RNN). Natural language processing (NLP) techniques were performed on textual clinical reports to structure the data and improve its usability for the model. Bag of words model and Term Frequency-Inverse Document Frequency (TF-IDF) vectorization were some of the few processes involved in extracting the prominent symptoms associated with each variant. The same has fed to various machine learning algorithms include multinomial naive Bayes (NB), logistic regression (LR), decision tree (DT), support vector machines (SVMs), and random forest (RF). Moreover, two more algorithms namely bagging, and gradient boost, are used for classification purposes. The reasons for choosing them are to observe their performance while classifying the proposed problem and compare the overall potency of the prediction system with RNNs. From this investigation, RNNs proved to be the optimal and expeditious technique to distinguish the covid strains. Adding more layers to the RNN has been noted to increase the efficacy of the system. The proposed approach with RNNs is giving an accuracy of around 99.14%.