Time Series Analysis of Infected COVID-19 Cases in the Zamboanga Peninsula, Philippines using Long Short-Term Memory Neural Networks

ABSTRACT

Infectious disease outbreaks, such as COVID-19 pandemics, exhibit patterns that can be described by the dynamics of a mathematical model This study seeks to explore the use of LSTM in order to develop models that will capture the non-linear dynamic changes of COVID-19 cases in Zamboanga Peninsula. The study uses 436 data points where the latest timestamp for the dataset is on May 29, 2021 and the oldest is on March 20, 2020. These data are taken from the DOH repositories and revalidated using the data from the DOH Regional Office. The training and testing phase results show that among the different LSTM variants, convLSTM trained using Adam and RMSProp attained the smallest RMSE result of 42.34 and 43.67 and a correlation coefficient of 0.94 0.93, respectively. ConvLSTM, when trained with Adam and RMSProp, produces the best results, as evidenced by the shortest RMSE and highest correlation coefficient. Results revealed that convLSTM appears to be a viable choice for modeling the time series of the COVID 19 infected cases in Zamboanga Peninsula Region in compared with the different variants of LSTM. © 2021 IEEE.