Leveraging the Machine Learning for Modelling the Respirable Coal Dust Composition: Evidence from the Zonguldak Coal Basin in Türkiye
Mining, Metallurgy and Exploration, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Basım Tarihi: 2026
- Doi Numarası: 10.1007/s42461-026-01725-7
- Dergi Adı: Mining, Metallurgy and Exploration
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus
- Anahtar Kelimeler: Coal mines, Machine learning, Modelling, Respirable coal dust
- Karadeniz Teknik Üniversitesi Adresli: Evet
Özet
Respirable coal dust is one of the most critical occupational health hazards in underground coal mines since prolonged exposure, particularly to respirable crystalline silica, is closely associated with coal workers’ pneumoconiosis (CWP) and silicosis. Therefore, accurate prediction of respirable dust composition is essential for effective exposure assessment, dust control, and occupational health management. This study presents the first machine learning (ML)-based framework for predicting the ash and quartz contents of respirable coal dust in the Zonguldak coal basin, the principal hard coal production region of Türkiye. A dataset comprising 56 independent field and laboratory observations collected from underground collieries was established using routinely measurable mining, ventilation, geological, and mineralogical parameters. The dataset was randomly divided into 85% training and 15% testing subsets, while model selection was performed exclusively on the training dataset using 5-fold cross-validation to improve methodological robustness and reduce overfitting. 27 ML-algorithms available in MATLAB environment were systematically evaluated. Predictive performances were assessed using the coefficient of determination (R²), root mean square error (RMSE), relative root mean square error (RRMSE), mean absolute percentage error (MAPE), and post error ratio (PER). Gaussian process regression (GPR) demonstrated the most consistent overall predictive performance, whereas support vector machine (SVM) and partial least squares regression (PLSR) also produced highly competitive results for quartz and ash prediction, respectively. The selected models achieved testing R² values ranging from 0.84 to 0.95 together with low prediction errors, indicating strong predictive performance within the investigated dataset. Pearson correlation analysis further enhanced model interpretability by identifying the variables contributing most strongly to ash and quartz prediction. Rather than proposing a new ML algorithm, this study demonstrates how established learning techniques can be effectively integrated with field and laboratory measurements to support respirable dust characterization, occupational health risk assessment, and proactive dust management in underground coal mines operating under geological and operational conditions similar to those investigated.