Application of Machine Learning Approaches for COVID-19 Spread Prediction in Ethiopia

Abstract

COVID-19 cases are increasing at an alarming rate all over the world. In Ethiopia, COVID-19 has infected hundreds of thousands and took thousands of lives. It is very important to know how many new cases and deaths to expect in the future, to be prepared in advance to avoid deaths from COVID-19. As a solution to this problem, this research proposed to build a prediction model for COVID-19 spread in Ethiopia using machine learning approaches. COVID-19 data was collected from the Federal Ministry of Health (FMOH), Ethiopia's official website. We utilized the data from March 13, 2020, to January 10, 2021. In this study, daily confirmed, death, and recovered COVID-19 cases of Ethiopia are modeled with Multi-Layer Perceptron (MLP), Auto Regressive Integrated Moving Average (ARIMA), Long-Short Term Memory (LSTM), and Facebook Prophet (FBProphet) approach. The performance Metrics used are Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Coefficient of Determination (R2 ). FBProphet model obtained the following results: for daily confirmed case RMSE=92.65, MAE=62.44, and R2 = 0.95. For daily death case RMSE=2.68, MAE=1.95, and R 2 = 0.83. For daily recovered case RMSE=75.67, MAE=44.69, and R2 = 0.98. According to the results of the first step of the study, FBProphet achieves slightly better performance than the other models, the MLP model achieves less performance in death and recovered cases, and the ARIMA model achieves less performance in confirmed cases. In the second stage of the study, the ARIMA model was provided to make predictions in a 7-day perspective that is yet to be known. Results of the second step of the study show that the total new confirmed and death cases decrease and Recovered cases increase in Ethiopia. Generally, COVID-19 spread prediction with the FBProphet machine learning method achieves a better performance than the ARIMA, LSTM, and MLP prediction models.