MODELLING THE PM2.5 CONCENTRATION WITH ARTIFICIAL INTELLIGENCE-BASED ENSEMBLE APPROACH

Abstract

Fine particulate matter (PM2.5) has been linked to a number of adverse health effects, hence its prediction for epidemiological studies has become very crucial. In this study, a novel ensemble technique was proposed for the prediction of PM2.5 concentration in cities with high traffic noise using traffic noise as an input parameter. Air pollutants concentration (P), meteorological parameters (M) and traffic data (T) simultaneously collected from seven sampling points in North Cyprus were used for conducting the study. The modelling was done in 2 scenarios. In scenario I, PM2.5 was modelled using 4 different input combination without traffic noise as input parameter while in scenario II, traffic noise was added as an input variable for 4 input combinations. The models were evaluated using 4 performance criteria including Nash-Sutcliffe Efficiency (NSE), Root Mean Square Error (RMSE), Correlation Coefficient (CC) and Bias (BIAS). Modelling PM2.5 with combined relevant input parameters of P, M and T could improve the performance of the model developed with only one set of the parameters by up to 12, 17 and 29% for models containing only P, M and T respectively. All the models in scenario II have demonstrated high prediction accuracy than the corresponding model in scenario I by up to 12% in the verification stage. The Support Vector Regression-based Ensemble model (SVR-E) could improve the performance accuracy of single models by up to 17% in the verification stage.