SO2 classification for air quality levels estimation using artificial intelligent techniques

This paper presents a new methodology to detect and classify SO₂ concentration according to the air quality level. In this classification, meteorological variables are analyzed to make a classification decision. The method consists of three steps. In first step, we group using a SOM neural networks the pollutant concentration in two classes, these classes are noise data and validated data. In second step, we create a representative feature vector using the information contingency levels that we know a priori. In third step, a new SOM neural network is trained with the representative feature vector built in second step, then the pollutant concentrations and meteorological variables (validated data) are self-organized in fourth classes according to contingency levels. Finally, we obtained the air quality level. Our experiments with this methodology exhibit good classification results. In this case a time series obtained from the environmental monitoring network of the Salamanca city, Guanajuato, Mexico is used.