Modelling of Air Pollution and Quality (Pm 2.5) in Bursa Urban Area via Machine Learning

In this study, it is aimed to model the meteorological conditions that play a role in the temporal distribution of air pollution parameter PM 2.5 by machine learning. For this purpose, daily PM 2.5 data were obtained from the database of the Ministry of Environment, Urbanisation and Climate Change between 2013-2015. In addition, in this study, a model was established between the daily maximum-minimum temperature obtained from the General Directorate of Meteorology and daily average temperature, pressure, wind speed and daily total precipitation. From this database, the days with no measurements were completed with missing data imputation. The database with missing data estimated was then made ready for the machine learning method. Support vector machine (SVM) is one of the most effective machine learning methods. These daily air pollution data and meteorological data were established using support vector regression radial kernel and then the best parameters were determined by applying tuning process. Accordingly, a correlation of 0.8 between the modelled and observed data and a Root Mean Square Error of 9.03 were obtained. This shows that the observation and modelling are quite consistent. In addition, the contribution of meteorological parameters in the model was determined. In this respect, pressure, minimum temperature, wind speed, average temperature, maximum temperature and precipitation are effective respectively. The effect of pressure conditions on the station, especially on the plain, is closely related to the inversion conditions.