Enforcement of Community Activity Restrictions Level Prediction in Jakarta Using Long Short-Term Memory Network

Abstract

The implementation of restrictions on community activities (Pemberlakuan Pembatasan Kegiatan Masyarakat – PPKM) is a strategy from the Indonesian government in handling the spread of COVID-19. PPKM is divided into four levels which will determine the restriction types that are to be implemented in a region. In this study, we aim to build a website that can predict PPKM levels through COVID-19 daily positive and death cases recorded in the Jakarta City, Indonesia. The prediction system uses the Long Short-Term Memory (LSTM) network and Node.JS as the backend of the website. We also introduced the usage of multivariate approach for this regression task by combining both daily positive and death cases number into the LSTM network. Based on the test scores obtained through evaluation using Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE), it was concluded that the proposed LSTM method could accurately predict the death cases with 0.17% MAPE and 22.68 RMSE but has poor performance in predicting the daily positive cases with 53.11% MAPE and 27.15 RMSE. This might be rooted from the use of multivariate approach during the model development where more variation to the daily positive cases was detected.