An empirical study of PM2.5 forecasting using neural network

In the recent years, a lot of efforts have been made to regulate air pollutant levels in most of the developed and developing countries. Fine particulate matter (PM2.5) is considered to be one of the major reasons behind deteriorating public health and a lot of efforts are being made to keep a check on PM2.5 levels. Accurately forecasting PM2.5 level is a challenging task and has been highly dependent on model based approaches. In this paper, we explore new possibilities to hourly forecast PM2.5. Choosing the right forecasting model becomes a very important aspect when it comes to improvement in prediction accuracy. We used Neural Network Autoregression (NNAR) method for the prediction task. The paper also provides a comparative analysis of prediction performance for additive version of Holt-Winters method, autoregressive integrated moving average (ARIMA) model and NNAR model. The experimentation and evaluation is done using real world measurement data from Airbox Project, which shows that our proposed method accurately does the prediction with significantly low error.