Predicting COVID-19 Cases using Deep LSTM and CNN Models

Abstract

The COVID-19 pandemic has had a profound and far-reaching impact on society. In order to effectively address this crisis, the timely implementation of necessary measures is crucial and accurate forecasting plays a vital role. In this context, this paper aims to use and compare deep learning techniques, specifically Long-Short Term Memory (LSTM) and Convolutional Neural Networks (CNN), for predicting the number of confirmed cases of COVID-19. To achieve this, the study examines the performance of CNN and LSTM architectures in forecasting the number of infected cases, both for one-day and seven-day predictions. Evaluation of these methods is based on the Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE) metrics, providing a comprehensive assessment of their effectiveness. The findings demonstrate that the CNN model proposed in this study exceeds the LSTM model, exhibiting superior prediction accuracy. Specifically, the CNN model achieves a mean MAPE score of 0.91 for one-day predictions and 4.85 for seven-day predictions, employing a ten-fold prediction time series split. These results highlight that both LSTM and CNN architectures are well-suited for forecasting tasks. The CNN model, in particular, shows excellent prediction efficiency, making it a promising approach for accurately estimating the number of cases of COVID-19 in the future.