Abstract: Air quality forecasting is one of the core elements of contemporary Urban Air Quality Management and Information Systems. Such systems are usually set up in order to serve environmental legislation needs and are tailored towards decision makers (for atmospheric quality problem abatement) and citizens (for early warning and information provision). The pluralism of forecasting methods that are available does not always lead to forecasting success, as the specific characteristics of each area of interest and the complicated, mostly chaotic relationships between air quality, meteorology, emissions and topography, limit the effectiveness of the methods used. On the other hand, the timescale of air quality problems dictate the usage of relatively `fast' methods, while the varying quality of input data calls for methods that have a low sensitivity in this factor and a high operational potential. For this reason, it is always interesting to perform a comparative study between various air quality forecasting methods and tools. The present paper describes the comparison work performed between several statistical methods and classification algorithms, on the basis of their performance to identify exceedances in the daily vegetation threshold. A second series of experiments is conducted for forecating of multiple hourly ozone values, where costs are empirically introduced for evaluating forecasting performance. Experiments are conducted on a dataset from the Marousi air quality monitoring station Athens, Greece.
I. N. Athanasiadis, K. D. Karatzas, P. Mitkas, Classification techniques for air quality forecasting, 5th ECAI Workshop on Binding Environmental Sciences and Artificial Intelligence (BESAI 2006), 17th European Conf on Artificial Intelligence, 2006.