Research Information System
Thesis Type: Postgraduate
Institution Of The Thesis: Gazi University, Fen Bilimleri Enstitüsü, Turkey
Approval Date: 2022
Thesis Language: Turkish
Student: BÜŞRA DUYGU ÇELİK
Supervisor: Nursal Arıcı
Open Archive Collection: AVESIS Open Access Collection
Abstract:
With the increase in energy use, urbanization, and heating activities that occur with the
increasing world population, air pollution has become a major environmental problem,
increasing day by day. The problem of air pollution causes serious damage to the ecosystem
by causing changes in air compounds in the atmosphere, acid rain, depletion of the ozone
layer, and negative effects on all living creatures. Estimating air pollution using a model is
crucial in taking measures for possible pollution and facilitating air quality improvement
processes. The fact that air pollution is dependent on many parameters, the inability to
measure all parameters and the high accuracy obtained at the end of the modelling are the
main reasons for using machine learning techniques in this regard. In this thesis, a data set
consisting of PM10, SO2, CO, NO2, NOX, O3 pollutants and meteorological parameters of
Temperature, Humidity, Precipitation, Wind Speed and Pressure measured between 2016-
2021 in Zonguldak and Manisa provinces are used. The data set was modelled with Support
Vector Machines, Decision Trees, Naive Bayes, Random Forest, K-Nearest Neighbor and
Long Short-Term Memory methods, which are well known in the literature. Model
performances measured and compared using accuracy, Root Mean Square Error (RMSE),
Mean Absolute Error (MAE), and Explanatory Coefficient (𝑅2) metrics. The LSTM, gives
better results than other methods with a 0.87 accuracy value for both stations. The lowest
performance is Naive Bayes with an accuracy value of 0.36.
Key Words : M achine learning, air pollution, air pollution prediction