Vertical roller mills (VRMs) dominate cement finish grinding because of their high efficiency and dry operation. However, the interaction among ventilation, fan operation, and pneumatic material transport remains poorly quantified using plant data. In this study, the previously established datadriven Conscious Lab (CL) framework, based on explainable machine learning, is developed to model material transport behavior in an industrial VRM circuit at the Tehran Cement Plant. A total of 1050 operating records collected over eight months under diverse conditions were cleaned and analyzed. Three ensemble algorithms (CatBoost, XGBoost, and Random Forest) were trained to predict two transport-relevant targets, mill fan speed and mill fan power, and were assessed with conventional hold-out splits and nested cross-validation to reduce selection bias. Among the evaluated models, CatBoost delivered the most reliable performance, achieving high predictive accuracy for fan speed (R2∼0.97) and fan power (R2∼0.95). Interpretability was ensured through SHAP analysis, which ranked the dominant drivers of each target and revealed nonlinear interactions among differential pressure, grinding pressures, water injection, and feed-related responses. The results highlight the important role of ventilation-related variables, particularly mill fan speed, fan power, and differential pressure, in governing pneumatic transport and circuit stability. The proposed CL framework provides an explainable, operator-oriented tool to support process diagnosis, setpoint decisions, and future development of real-time intelligent control strategies for energy-efficient cement grinding.