COVID-19 Time-Varying Reproduction Number and its Prediction by Adaptive Neuro-Fuzzy Inference System (ANFIS)

Abstract

Epidemiologists and the COVID-19 Response Acceleration Task Force in Jakarta, Indonesia continue to monitor various parameters of the outbreak in Jakarta and its surroundings. One of the parameters to monitor and predict is the effective reproduction number (Rt ). We propose a machine learning approach called adaptive neuro-fuzzy inference system (ANFIS) to predict the time-varying Rt . We calculated historical data for Rt using the daily confirmed cases, daily death cases, and daily recovered cases by applying a Susceptible-Infected-Recovered-Dead (SIRD) dynamic simulation. We found that using the input data from the last seven previous days of Rt as the training data gave best prediction results based on the evaluation metric of mean absolute percentage error (MAPE). We validated the prediction results with 10 consecutive days of Rt from November 1–10, 2020. The results showed that the developed ANFIS model could predict the reproduction number for COVID-19 in Jakarta, Indonesia with an accuracy of approximately 81.27%. The transmission of COVID-19 in Jakarta, Indonesia has the potential to be controlled because most Rt values were lower than 1, at least during the period analyzed in this study.Funding: None to declare. Declaration of Interest: None to declare.