Prediction of Air Pollutant Concentrations in Airport Areas using Machine Learning Architecture

The importance of analyzing and predicting the levels of atmospheric pollutants is emphasized in environmental risk studies. Existing prediction models based on machine learning have been proposed for this purpose. In the case of experimental campaigns conducted in airport areas, one such pollutant is Sulphur HexaFluoride (SF
6
), a gas that is odorless, colorless, and rare in the air. Due to the weak linear relationship between atmospheric variables and the concentration of SF
6
, linear models are not suitable for this prediction task. To address this challenge and accurately predict atmospheric pollutant concentrations in airport areas, two prediction models were chosen and developed: a random forest machine learning model and a CNN-LSTM architecture, which is a deep learning approach. A comparison of results obtained by the two models achieved on two campaigns measures (Capitoul & Indianapolis) is presented. The accuracy of the prediction results is evaluated using the coefficient of determination (R
²
) and the Mean Absolute Error (MAE) as metrics.