A comparative study of LSTM, Bi-LSTM, and DBi-LSTM network model in forecasting COVID-19 new cases and new deaths in Indonesia

The Coronavirus Disease 2019 (COVID-19) pandemic has become one of the biggest challenge in research community, especially in public health system. One of the challenges that encountered is how to manage the public health system so that it does not collapse because of overcapacity, while maintaining the testing, tracing, and treatment to slowing down the spread of the virus and decreasing the number of deaths caused by COVID-19. Therefore, a predictive model that can estimate and forecast the new daily cases and daily deaths of COVID-19 are important to give recommendations to policy maker about how to control - and eventually - prevent the spread. In this paper, we present a comparative study between Long short-term memory (LSTM), Bidirectional LSTM (Bi-LSTM), and Deep Bidirectional LSTM (DBi-LSTM) to forecast the number of COVID-19 daily cases and daily deaths in Indonesia, based on daily confirmed cases and daily death cases from August 26th, 2020 to August 25th, 2021. The result is DBi-LSTM performs better than LSTM and Bi-LSTM in both daily cases and daily deaths datasets, in terms of training process and lower error prediction value based on MAE and RMSE. Here we also provide 14 day out-of-sample forecast for daily cases and daily deaths from August 26th to September 8th, 2021 using each model.