Classification of Zinc Recovery Quality from EAF Dust Using Machine Learning: A Waelz Process Study

Zinc recovery from Electric Arc Furnace (EAF) dust represents a significant challenge in the iron and steel industry. This study aims to classify zinc quality in slag produced through the Waelz process, where zinc is reduced and volatilized at high temperatures (>1000°C) in rotary kilns, using machine learning techniques. The classification of zinc quality in slag is crucial for process optimization and environmental sustainability, as it directly impacts both resource recovery efficiency and waste management strategies. The dataset utilized for developing classification models was obtained from chemical analyses of Waelz process raw materials and slag samples. Four distinct classification algorithms (Support Vector Machine SVM, Decision Tree - DT, Naive Bayes - NB, and Random Forest - RF) were evaluated on the data labeled by experts according to zinc content in slag. The reliability of the models was assessed through 10-fold cross-validation. In experimental studies, the DT algorithm demonstrated superior performance with 100.0% accuracy, precision, sensitivity, and F1 score. The RF algorithm achieved second-place performance with 96.0-98.0% accuracy and 100.0% precision, followed byNB with 91.0-94.0% accuracy, and SVM with 84.0-88.0% accuracy. The results indicate that the DT algorithm can serve as a reliable tool for quality classification in the zinc re-covery process. These findings contribute significantly to the advancement of automated quality control systems in metallurgical processes, potentially enabling real-time monitoring and optimization of zinc recovery operations.