Forecasting the Condition Severity of COVID-19 in Terms of Prevalence in Iran

Since the advent of the Coronavirus disease 2019 (COVID-19) pandemic, one of the most significant attempts toward coping with this disease was predicting the severity condition of this pandemic in terms of prevalence. In this way, the authorities can make better and more informative decisions. Machine Learning has great potential to solve such data-driven problems. In this article, the issue of predicting the future COVID-19 prevalence severity as a classification task has been investigated. We employed four state-of-the-art predictive models, Support Vector Machine (SVM), Multi-Layer Perceptron (MLP), Long Short-Term Memory (LSTM), and Self-Attention mechanism, in order to predict the COVID-19 prevalence severity in Iran using the gathered dataset by Johns Hopkins University. The main challenge we have faced was data scarcity because of the limited duration of the pandemic. Besides, as we are considering a classification task, data imbalance was an issue due to imposing a conventional criterion to divide each day into three classes regarding the state of severity. The final results suggest that although the performance of all models was acceptable, Self-Attention with an accuracy of 95% was the superb architecture to deal with this classification task.