Data mining methods for prediction of air pollution

Abstract

The paper discusses the methods of data mining for prediction of air pollution. Two problems in such prediction are important: the generation and selection of the prognostic features, and final prognosis of the pollution level for the next day on the basis of the data of the previous day. In this paper we analyze and compare two methods of feature selection. One applies the genetic algorithm, and the second the linear method of stepwise fit. On the basis of such analysis we are able to select the most important features influencing the prediction. As a mathematical tool for final prediction we apply the neural networks. Three different solutions will be compared: the multilayer perceptron (MLP), radial basis function (RBF) network and support vector machine (SVM).