Predicting Covid-19 Cases for 12 Countries using Long Short-Term Memory

Abstract

A novel virus named coronavirus or ‘COVID-19’ by the World Health organization (WHO) has spread around the entire world placing mankind in a situation that no one had predicted. The rise of the number of infected and death cases around the world is alarming and has caused hysteria among mankind. Considering the adversity of the COVID-19, some immediate plan to monitor the number of cases in the future needs to be maneuvered. In this paper, we aim to implement a method to envision the number of COVID-19 cases for the future. We achieve the result by using a deep learning algorithm known as Long Short-Term Memory (LSTM) over the real-time dataset provided by WHO for predicting the number of COVID-19 cases in twelve countries. The countries considered in this study are United States of America, China, United Arab Emirates, India, Brazil, France, Germany, Spain, Republic of Korea, Italy, Singapore, and Argentina. The contribution of this paper is to provide each country with their own model that can help predict their respective future COVID-19 cases. With these predictions, each country can then come up with solutions to reduce the number of infected cases in their respective nation. The proposed LSTM model was evaluated using metrics such as Correlation Coefficient and R2 Error. The results show that the model was giving high R2 score (≥ 0.7) and high correlation coefficient (≥ 0.7) between the test and train datasets. In the cases where R2 score (< 0.7) and correlation coefficient (< 0.7) were low, the train and test values of the datasets were similar making the predictions accurate.