Forecasting of COVID-19 Cases in Jakarta using Poisson Autoregression

Abstract

COVID-19 is currently become a global problem, including in Jakarta, Indonesia. There have been many approaches to predict COVID-19 occurrence, including the forecasting approach. However, the traditional forecasting method, particularly machine learning, often does not consider the condition of the data, although it has forms of the count, such as the number of cases. This study employs an autoregression model using Poisson distribution in predicting the COVID-19 future cases, namely the positive and recovery number. We compare the Poisson Autoregression with several well-known forecasting methods, namely ARIMA, Exponential Smoothing, BATS, and Prophet. This study found that Poisson Autoregression could create an accurate prediction with MAPE below 20% and tend to follows the actual data for the next 8 to 14 days to the future. Thus, this approach can forecast the future cases of COVID-19 and other cases that use count data in Jakarta, like the number of citizen complaints or transportation context.