An Effective Hybrid Model for Predicting Air Quality of Ankara

Abstract

Increasing industrialization, population growth, urbanization and increase in fossil fuel consumption lead to air pollution that affects human health by polluting the atmosphere. Particulate matter, known as PM10 and PM2.5, are air pollutants that can remain suspended in the air in solid, liquid or both states. Substances are described according to their aerodynamic diameter, known as particle size. Estimating particulate matter concentrations is very important for human health and the environment. In this study, a hybrid deep learning model was developed for air quality prediction using PM2.5 and PM10 concentration data obtained from Bahçelievler, Demetevler, Sincan and Törekent air quality monitoring stations in Ankara. In the developed model, it was aimed to use the successful features of CNN and LSTM models. The developed CNN-LSTM model was compared with LR, RF, SVM, MLP, CNN and LSTM using MSE, RMSE, R2, and MAE. Experimental results showed that the CNN-LSTM model outperformed the compared models and each station had an R2 of approximately 0.9.