Predicting COVID-19 trends: a deep dive into time-dependent SIRSD with deep learning technique / Abdul Basit ... [et al.]

The COVID-19 pandemic, also known as Coronavirus Disease 2019, has affected over 700 million people globally, resulting in approximately 7 million deaths. Research has proposed multiple mathematical models to institute a disease transmission framework and predict the disease growth. Most of the existing mathematical disease growth prediction models are less effective due to the exclusion of the re-susceptible scenarios and overlooks their time-dependent properties, which change continuously during the viral transmission process. Another popular prediction technique is deep learning approaches. However, existing methods often fail to accurately capture the dynamic trends of epidemics during their spreading phases in short-term and medium term. Therefore, inspired by the deep learning approach, this study offers a new model for COVID-19 prediction centered on time-dependent namely Susceptible-Infected-Recovered-re-Susceptible-Death-Deep Learning (SIRSD-DL) model. This model proposes a combination of deep learning techniques, specifically Feed-Forward Neural Networks (FFNN) and Recurrent Neural Networks (RNN), with an epidemiological mathematical framework. It aims to forecast the parameters of SIRSD model by incorporating deep learning technology With the current COVID-19, we examined data from seven countries—China, Malaysia, India, Pakistan, South Korea, the United Arab Emirates and the United States of America—between March 15, 2020, till May 27, 2021. Our research demonstrates that the proposed model outperforms both standalone and hybrid techniques, offering enhanced predictability for short- and medium-term forecasts. In India, the model achieved prediction accuracies by Mean Absolute Percentage Error of 0.82% for 1-day, 1.48% for 3-day, 2.72% for 7-day, 2.50% for 14-day, 3.73% for 21-day, and 6.63% for 28-day forecasts. This approach is expected to be