Prediction of COVID-19 by Its Variants using Multivariate Data-driven Deep Learning Models

TL;DR

This study employs deep learning models, including LSTM, BiLSTM, and RNN, to predict COVID-19 confirmed cases based on variants using European dataset, achieving high accuracy in variant-specific forecasts.

Contribution

It introduces a comparative analysis of LSTM, BiLSTM, and RNN models for COVID-19 case prediction based on variants, highlighting the effectiveness of BiLSTM and RNN in this context.

Findings

RNN outperformed other models with lowest MSE and RMSE.

BiLSTM achieved the best results with specific layer and hidden sizes.

Models accurately predicted cases for specific variants like B.1.427/B.1.429.

Abstract

The Coronavirus Disease 2019 or the COVID-19 pandemic has swept almost all parts of the world since the first case was found in Wuhan, China, in December 2019. With the increasing number of COVID-19 cases in the world, SARS-CoV-2 has mutated into various variants. Given the increasingly dangerous conditions of the pandemic, it is crucial to know when the pandemic will stop by predicting confirmed cases of COVID-19. Therefore, many studies have raised COVID-19 as a case study to overcome the ongoing pandemic using the Deep Learning method, namely LSTM, with reasonably accurate results and small error values. LSTM training is used to predict confirmed cases of COVID-19 based on variants that have been identified using ECDC's COVID-19 dataset containing confirmed cases of COVID-19 that have been identified from 30 countries in Europe. Tests were conducted using the LSTM and BiLSTM models with the addition of RNN as comparisons on hidden size and layer size. The obtained result showed that in testing hidden sizes 25, 50, 75 to 100, the RNN model provided better results, with the minimum MSE value of 0.01 and the RMSE value of 0.012 for B.1.427/B.1.429 variant with hidden size 100. In further testing of layer sizes 2, 3, 4, and 5, the result shows that the BiLSTM model provided better results, with minimum MSE value of 0.01 and the RMSE of 0.01 for the B.1.427/B.1.429 variant with hidden size 100 and layer size 2.