Forecasting Time Series with Multiplicative Trend Exponential Smoothing and LSTM: COVID-19 Case Study

Arceda M.A.M.; Laura P.C.L.; Arceda V.E.M. · Advances in Intelligent Systems and Computing (2021) · DOI: 10.1007/978-3-030-63089-8_36

Resumen

In this work, we present an analysis of time series of COVID-19 confirmed cases with Multiplicative Trend Exponential Smoothing (MTES) and Long Short-Term Memory (LSTM). We evaluated the results utilizing COVID-19 confirmed cases data from countries with higher indices as the United States (US), Italy, Spain, and other countries that presumably have stopped the virus, like China, New Zealand, and Australia. Additionally, we used data from a Git repository which is daily updated, when we did the experiments we used data up to April 28th. We used 80% of data to train both models and then, we computed the Root Mean Square Error (RMSE) of test ground true data and predictions. In our experiments, MTES outperformed LSTM, we believe it is caused by a lack of historical data and the particular behavior of each country. To conclude, we performed a forecasting of new COVID-19 confirmed cases using MTES with 10 days ahead.

Instituciones: Universidad Nacional de San Agustín de Arequipa; Universidad La Salle