Air Pollution Prediction Using Machine Learning Algorithms

Air pollution is a global environmental problem that poses serious threats to human health and quality of life, especially in industrializing and densely populated areas. The use of machine learning techniques in air pollution prediction has become an important tool in air quality management. In this study, PM10 concentrations were estimated using support vector machines (SVM), artificial neural networks (ANN), convolutional neural networks (CNN), random forest (RF), gradient boosting (GB), K ‐nearest neighbor (KNN), linear regression (Lasso), and long short‐term memory (LSTM) models, and the models were examined in detail. The results showed that the SVM model performed best with the lowest error values (mean absolute error [MAE]: 4.747 µg/m 3 and root mean square error [RMSE]: 6.648 µg/m 3 ), whereas the ANN model achieved the highest accuracy level ( R 2 : 0.791). The KNN model showed the weakest performance with the highest error rates (RMSE: 10.07 µg/m 3 ). Among the LSTM models, the model trained with the Adam optimization algorithm exhibited the best performance compared to other LSTM versions. Among the models used in PM10 estimation, SVM, ANN, and CNN highlight the effectiveness of machine learning models for air pollution estimation and constitute an important reference for studies to be conducted in this field. Focusing on Karaman, a medium‐sized industrial city dominated by food‐processing industries, it provides a representative framework for similar medium‐sized industrial cities facing comparable environmental challenges. The research introduces an integrated and transparent machine‐learning approach that combines data preprocessing, outlier correction, normalization, and broad hyperparameter optimization under consistent conditions for eight algorithms. This framework improves the reliability, comparability, and reproducibility of air‐quality prediction.