Indirect Estimation of Carbon Emissions in Cement Plants Using Machine Learning: A Random Forest Approach for Alternative Biomass Co-Processing

(1) Background: The cement industry is a major contributor to global greenhouse gas emissions, requiring reliable monitoring methodologies. This study addresses the indirect estimation of combustion-related CO2 associated with fuel consumption in a clinker kiln from an operational cement plant, where a complete direct hourly fuel-emissions record was unavailable. (2) Methods: Industrial SCADA data were preprocessed to address missing values, harmonize hourly intervals, and exclude shutdown periods. Three independent random forest models were developed for petroleum coke, olive pomace, and grape pomace, and the predicted fuel flows were converted into physical combustion CO2 using fuel-specific emission factors. Model performance was assessed through a random hold-out evaluation and a complementary chronological sensitivity analysis designed to examine temporal transferability. The analysis focuses on the performance and interpretation of the modular random forest architecture rather than on comparative algorithm selection. (3) Results: Under the random hold-out evaluation, the three models achieved R2 values between 0.840 and 0.851. The chronological analysis yielded lower and fuel-dependent performance, indicating that temporal changes in operating conditions affect model transferability. The random-split results therefore characterize interpolation within the observed operating envelope, whereas the chronological analysis provides a more demanding assessment on later operating periods. (4) Conclusions: The modular framework provides a proof of concept for the indirect estimation of fuel-related combustion CO2 under the operating conditions represented in the dataset. Application to future periods requires temporal validation, drift monitoring, and periodic retraining. The framework does not estimate calcination CO2, replace certified CEMS, or establish superiority over other regression families.