Random forest-based waste prediction in sustainable graphic production within industry 4.0

Graphic production systems generate significant material waste due to de fective prints, registration errors, inefficient material utilization, and subop timal energy management. In printing processes, waste levels can exceed 10%, contributing to environmental burden and increased operational costs. Within Industry 4.0 environments and ESG-oriented production management, companies require measurable, data-driven mechanisms for resource optimization. The aim of this research is to develop and evaluate a Random Forest-based predictive model for production waste forecasting in sustainable graphic systems using key performance indicators as opera tional inputs.A supervised Random Forest regression model was developed to predict waste levels. Because publicly accessible industrial datasets are unavailable due to confidentiality constraints, simulated KPI data was generated to re flect realistic production variability. Input variables include material waste percentage, machine efficiency, number of defective prints, energy con sumption, and equipment downtime. The synthetic dataset incorporates controlled variability and inter-variable correlations to approximate real production conditions. The data was divided into training and testing sub sets, and cross-validation was applied to ensure model robustness. Model performance was evaluated using mean absolute error, mean squared er ror, and the coefficient of determination.The Random Forest model achieved an R² value of 0.89, indicating strong explanatory power, with an overall prediction accuracy of 87.32% and a mean absolute error of 3.17%. Simulation scenarios demonstrate substan tial waste reduction potential under AI-supported optimization. In the printing process, waste levels decreased from 12.7% to 5.8%, represent ing a 54.3% reduction. Early detection mechanisms identified 36% of po tential waste events prior to occurrence, enabling proactive intervention. Additional simulation results indicate projected improvements in energy efficiency, defect prevention, and overall cost reduction, estimated at 8%, 12%, and 6%, respectively.The findings confirm that Random Forest-based predictive modeling can function as an operational tool for ESG-aligned waste management in In dustry 4.0 graphic production. KPI-driven monitoring combined with pre dictive analytics enables early identification of inefficiencies, improved pro duction planning, and systematic reduction of material losses. The model is suitable for integration with sensor systems and production management platforms in smart manufacturing environments, supporting both environ mental sustainability and economic performance.