Modeling of PM₁₀ concentrations in Erzurum's Aziziye district using artificial neural networks

In this study, it was aimed to model the PM₁₀ concentration in the Aziziye district of Erzurum province using air quality and meteorological data from the year 2024. The data were obtained as daily averages from the Aziziye Air Quality Monitoring and Erzurum Meteorology Stations. The main objective of the study is to develop highly accurate models that can estimate PM₁₀ concentrations, which require specialized instruments and infrastructure for direct measurement, through meteorological and gaseous pollutant parameters. In this context, Artificial Neural Network (ANN) algorithms and DataFit software were used to construct linear and nonlinear regression models, and the predictive performances of both methods were compared using statistical indicators. During the parameter selection process, t-ratio analysis, Pearson correlation coefficient, and Principal Component Analysis (PCA) methods were utilized; final variables were determined based on environmental knowledge, literature, and modeling experience. ANN models were evaluated through training, validation, and testing datasets, and the best-performing model yielded R² = 0.9607 and MSE = 0.00117 for the test data. The multiple linear regression model developed with DataFit achieved an R² value of 0.89. The results demonstrated that both models provided high accuracy and statistical reliability. Furthermore, it was evaluated that the developed model structures have the potential to support regional air quality management processes when appropriately adapted. In this respect, the study provides not only a reliable PM₁₀ prediction approach but also a methodological and regional contribution to similar environmental modeling research.