University Of Tehran Press Pollution 2383-451X 11 4 2025 10 01 A Hybrid Machine Learning Model Based on Deep Learning for Air Quality Prediction 1199 1215 103797 10.22059/poll.2025.388743.2750 EN Mohammad Reza Mehregan Department of Industrial Management, Faculty of Industrial Management and Technology, College of Management, University of Tehran, Tehran, Iran Mohammad Taghi Taghavifard Department of Information Technology and Operations Management, Faculty of Management and Accounting, Allameh Tabataba'i University, Tehran, Iran Amir Mohammad Khani Department of Industrial Management, Faculty of Industrial Management and Technology, College of Management, University of Tehran, Tehran, Iran Arman Rezasoltani Department of Industrial Management, Faculty of Industrial Management and Technology, College of Management, University of Tehran, Tehran, Iran Mohammad Ali Nikkhah Department of Industrial Management, Faculty of Industrial Management and Technology, College of Management, University of Tehran, Tehran, Iran Journal Article 2025 01 18 Air pollution is a major global challenge, significantly and directly affecting public health, urban sustainability, and environmental policy. Accurate air quality prediction has increasingly become essential to address the challenges posed by environmental adversities. This study proposes a novel hybrid machine learning model that combines deep learning and advanced ensemble techniques to improve air quality prediction. This model combines Deep Neural Network (DNN), along with ensemble learning algorithms such as XGBoost, CatBoost, LightGBM, and Random Forest as a metamodel to aggregate the predictions. The model was tested on a dataset that included environmental aspects ranging from PM2.5, PM10, CO, and NO2 variables to socio-economic variables such as proximity to industrial areas and population density. Feature selection and data imbalance were handled using RFECV and SMOTE, respectively. The tuning of the hyperparameters in the model was done using both TPE implemented by Optuna and Bayesian optimization by Keras-Tuner. This model can achieve a remarkable accuracy of 97.34%, which is superior to conventional approaches. The results present a case for building hybrid machine learning techniques for air quality prediction as a basis for intelligent global environmental monitoring in an interpretable, accurate, and scalable manner. Future work can integrate the real-time incoming data from the Internet of Things (IOT) and extend the model concept for multi-prediction benchmarks to other environmental indices, thus broadening its horizon and applicability to upcoming global environmental challenges. https://jpoll.ut.ac.ir/article_103797_d56e8d34a1239b7ee63de66ef7e275d2.pdf