Forecasting Air Pollutant Index (API) Using Nonlinear Autoregressive (NAR) Neural Network During Covid-19 Pandemics in Malaysia

Abstract: Coronavirus Disease 2019 (COVID-19) pandemics have emerged in Malaysia since 18 March 2020, which then the government has announced for Movement Control Order (MCO) as a method to curb the transmission in public. The air quality is expected to be good as most of the operations are closed. Thus, we evaluated and predicted the Air Pollutant Index (API) during the MCO in Malaysia for an overview of the air quality level during the pandemic. As the API is complex in the atmosphere, we used a nonlinear autoregressive (NAR) neural network model for the nonlinear dataset. Urban cities are generally having higher pollutants concentrations along with the urbanization process. High pollutant concentrations led to health problems, especially respiratory illness, either in the short or long term. We used the data from 18 March 2020 (the first day of Movement Control Order, MCO) until 31 December 2020. Results revealed the NAR model executed higher R2 for Kuala Terengganu (99.23%). The optimum NAR model architectures which are trained using the Levernberg-Marquardt training algorithm is 1:14:1 for Kuala Terengganu. NAR neural network is capable of modeling and forecasting nonlinear time series during the COVID-19 pandemic.