Accurately estimating ink consumption in the packaging printing industry is crucial for minimising waste, improving efficiency, and promoting sustainability. Traditional estimation methods rely on operator intuition, often leading to overproduction and increased costs. This study applies machine learning to predict pre-printing ink consumption using a dataset of 77 samples from a domestic packaging printing company. Key numerical and categorical features, including paper size, halftone dot percentage, ink type, and paper properties, were analysed. Multiple machine learning models were tested, with XGBoost outperforming others, achieving a mean absolute error (MAE) of 0.0026 g, root mean squared error (RMSE) of 0.0035 g, and an R-squared value of 0.619. Implementing XGBoost led to an average ink savings of 2,380 g per job. These findings highlight the potential of machine learning, particularly XGBoost, in optimising ink consumption. Future research will explore additional paper properties and customised objective functions to enhance prediction accuracy.
Balcioglu et al. (Wed,) studied this question.