Modeling Dust Emissions Using ANN, XGBoost, and Random Forest Techniques: A Case Study of the Meftah Cement Plant (Algeria)

Abstract

The production of clinker in cement plants remains a significant source of atmospheric dust emissions, posing a major challenge for compliance with environmental standards and the protection of public health. This research evaluates and compares the performance of three machine learning architectures—Artificial Neural Networks (ANN), XGBoost, and Random Forest—to predict dust levels from the AFF2 stack at the Meftah cement plant (Blida, Algeria) in near real-time. The predictive models are developed based on five key operational variables: burned gas rate, exhaust gas temperature at the tower outlet, raw meal feed rate (flour), filter differential pressure (AFF2), and excess air percentage (O2). For model training and validation, an exhaustive dataset comprising over 13,000 historical observations from 2023 and 2024 was utilized. The obtained results demonstrated that ensemble tree-based models significantly outperformed the ANN model. Among the investigated approaches, the Random Forest model achieved the best predictive performance with the lowest error values (MSE = 8.854, RMSE = 2.976, and MAE = 1.941) and the highest coefficient of determination (R² = 0.750). The XGBoost model also produced strong predictive capability with an R² value of 0.736, while the ANN model showed comparatively lower performance with an R² value of 0.645. Residual analysis and prediction-versus-actual plots further confirmed the superior robustness and generalization capability of the Random Forest algorithm. The findings demonstrate the effectiveness of machine learning techniques for real-time prediction of dust emissions in cement manufacturing processes. The proposed predictive framework can serve as an intelligent decision-support tool for environmental monitoring, predictive maintenance, and proactive control of particulate emissions. The implementation of such models may help cement plants reduce environmental impacts, optimize filtration system performance, and improve compliance with environmental regulations.