Urban Air Quality Forecasting: A Regression and a Classification Approach

We employ Computational Intelligence (CI) methods to model air pollution for the Greater Gdansk Area in Poland. The forecasting problem is addressed with both classification and regression algorithms. In addition, we present an ensemble method that allows for the use of a single Artificial Neural Network-based model for the whole area of interest. Results indicate good model performance with a correlation coefficient between forecasts and measurements for the hourly PM10 concentration 24 h in advance reaching 0.81 and an agreement index (Cohen's kappa) up to 54%. Moreover, the ensemble model demonstrates a decrease in Mean Square Error in comparison to the best simple model. Overall results suggest that the specific modelling approach can support the provision of air quality forecasts at an operational basis.