Machine Learning-Based Demand Forecasting in the Packaging Industry: A Random Forest Approach

Authors:

Abstract
This study focuses on demand forecasting for packaging products, specifically drums and cans, employing the Random Forest Regressor (RFR) as a data-driven approach to predict packaging demand. It investigates the impact of data split scenarios and hyperparameter tuning on the forecasting testing data split ratio that produces optimal model performance. The study aims to identify the best Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE) values achieved through optimized hyperparameter configurations, and to evaluate the forecasting capability of the Random Forest model in supporting operational decisionmaking for packaging inventory management. The methodology involves an experimental design with multiple data split scenarios and implements systematic hyperparameter tuning, involving the number of trees, maximum depth, and minimum samples per split. The results show that the 80:20 data split gave the best balance between learning and generalization, which is in line with previous findings in the literature. After hyperparameter optimization, the Random Forest model achieved a MAPE of $\mathbf{8. 4 4 \%}$ and an RMSE of $\mathbf{1, 9 5 9}$. This shows a significant improvement in forecast accuracy compared to the baseline configurations. This study contributes valuable insights into implementing Random Forest Regressor techniques for demand forecasting in the packaging industries, highlighting the importance of a comprehensive data split for training and validation, as well as hyperparameter tuning evaluation, to enhance forecast quality and inform operational decisionmaking in inventory management.

This publication has 10 references indexed in Scilit: