Studies of air quality predictors based on neural networks

In recent years, urban air pollution has emerged as an acute problem because of its negative effect on health and living conditions. Regional air quality problems, in general, are linked to violations of specified air quality standards. The current study aims to find neural network based air quality predictors, which can work with a limited number of datasets and are robust enough to handle data with noise and errors. A number of available variations of neural network models, such as the Recurrent Network Model (RNM), the Change Point Detection Model with RNM (CPDM), the Sequential Network Construction Model (SNCM), the Self Organizing Feature Model (SOFM), and the Moving Window Model (MWM), were implemented using MATLAB software for predicting air quality. Developed models were run to simulate and forecast based on the annual average data for 15 years from 1985 to 1999 for seven parameters, viz. VOC, NOx, CO, SO2, PM10, PM2.5 and NH3 for one county of California, USA. The models were fitted with first nine years of data to predict data for remaining six years. The models, in general, could predict air quality patterns with modest accuracy. However, the SOFM model performed extremely well in comparison with the other models for predicting long-term (annual) data.