A prediction distribution of atmospheric pollutants using support vector machines, discriminant analysis and mapping tools (Case study: Tunisia)

Abstract

Monitoring and controlling air quality parameters form an important subject of atmospheric and environmental research today due to the health impacts caused by the different pollutants present in the urban areas. The support vector machine (SVM), as a supervised learning analysis method, is considered an effective statistical tool for the prediction and analysis of air quality. The work presented here examines the feasibility of applying the SVM to predict the ozone and particle concentrations in two Tunisian cities, namely Tunis and Sfax. We used the SVM with the linear kernel, SVM with the polynomial kernel and SVM with the RBF kernel to predict the ozone and particle concentrations in Tunisia for one year. The RBF kernel produced good results for the two pollutants with 0% error rate. Polynomial and linear kernels produced sufficiently low errors for the pollutants, at 9.09% and 18.18%, respectively. Discriminant Analysis (DA) was selected to analyze the datasets of two air quality parameters, namely ozone O3 and Suspended Particles SP. The DA results show that the spatial characterization allows for the successful discrimination between the two cities with an error rate of 4.35% in the case of the linear DA and 0% in the case of the quadratic DA. A thematic map of Tunisia was created using the MapInfo software.