Artificial neural network model for predicting air pollution. Case study of the moravica district, Serbia

Abstract

An example of artificial neural network model for predicting air pollution has been presented. The research was conducted in Serbia, the Moravica District, on the territory of two municipalities (LuCani and Ivanjica) and the town Čačak. The level of air pollution was classified by a neural network model according to the input data: municipality, site, year, levels of soot, sulfur dioxide (SO2), nitrogen dioxide (NO2) and particulate matter. The model was evaluated using a lift chart and a root mean square error (RMSE) has been determined, whose value was 0.0635. A multilayer perceptron has also been created and trained with a back propagation algorithm. The neural network was tested with the data mining extensions (DMX) queries. The results have been obtained for air pollution based on new input data that can be used to predict the level of pollution in future if new measurements are carried out. A web-based application was designed for displaying the results.