Air Quality Prediction: An Opportunistic Neuro-Ensemble Approach

Abstract

The present article discusses the development of neural-network-based air quality prediction models which can work with a limited number of data sets and are robust enough to handle data with noise. Five different variations of neural network models (partial recurrent network (PRNM), sequential network construction (SNCM), self-organizing feature maps (SOFM), moving window (MWM), and integrated normalized autoregressive moving average-self-organized feature maps models (NARMA-SOFM)), were implemented in a WINDOWS environment using MATLAB software. Developed models were run to simulate and forecast the daily average data for three parameters: RPM (respirable particulate matter), SO2 (sulphur dioxide), and NO2 (nitrogen dioxide) for the Ashram Chowk location in New Delhi, India. The implemented models were found to predict air quality patterns with modest accuracy. To improve the models’ performance, an innovative approach using an opportunistic ensemble of the first four developed neural network models (OEM) was proposed for predicting the same short-term data. The ensemble approach indeed demonstrated an improvement on earlier models. However, the NARMA-SOFM model performed the best.