Improving COVID-19 Forecasts using Hybrid Machine Learning Models

The spread of COVID-19 has impacted human life in every aspect. Because of the uncertainty of the virus’s future impact, highly advanced machine learning approaches are needed to enhance the accuracy of COVID-19 forecasts. Therefore, we propose the following hybrid models to address the challenge: CNN-LSTM-Prophet, GRU-LSTM-Prophet, and RNN-LSTM-Prophet. The RNN-LSTM-Prophet model performs the best in predicting daily COVID-19 cases in the United States as it has the lowest MSE, RMSE, and MAE values. Moreover, all of the hybrid models perform better than the individual LSTM and Prophet models. Based on the benchmark results, RNN-LSTM-Prophet performs the best compared to the other models in this study. The hybrid model also performs better than the baseline as the MSE, RMSE, and MAE are lower by 2.48%, 1.24%, and 1.85% respectively. Governments worldwide can use RNN-LSTM-Prophet to better determine COVID-19’s future trajectory and then implement the appropriate policies and protocols. Moreover, the model can be used to deal with new diseases on the rise.