The Role of Data, Machine Learning, and Supply Chain Interdependencies in Implementing Connected Packaging Solutions for Enhanced Supply Chain Efficiency

Abstract

The complex SCM method combines several elements, which include demand forecasting alongside inventory management and supplier selection, and risk mitigation procedures. Artificial Intelligence integration with Supply Chain Management shows the potential to boost operational effectiveness while upgrading decision systems and minimization of operational safety threats. The study aims to investigate and apply AI methods to SCM for efficiency improvement, decision-making aids, and operational risk mitigation, using novel computational tools such as predictive and optimization models. This study proposes an integrated methodology for demand forecasting, late delivery risk prediction, and delivery status classification to enhance decision-making in SCM. The "DataCo Smart Supply Chain for Big Data Analysis"(DataCo Smart Supply Chain for Big Data Analysis Dataset, no date) dataset is used, which is sourced from Kaggle. The key preprocessing steps included missing data handling, datetime transformation, feature engineering, and variable encoding. Time series forecasting using Prophet was applied to weekly aggregated order data, while classification tasks utilized AdaBoost, Cat Boost, and MLP models. Feature importance analysis and SMOTE were employed to enhance model accuracy and address data imbalance. Model performance was evaluated using MAE, MSE, RMSE, accuracy, recall, precision, and F1 score. The proposed machine learning models demonstrated high effectiveness in optimizing key supply chain tasks. For delivery status prediction, advanced models such as AdaBoost, Cat Boost, and MLP achieved exceptional AUC scores, with Cat Boost and MLP reaching 100%. In late delivery risk prediction, these models consistently delivered strong performance, achieving AUC values above 97%. For demand forecasting, the Prophet model achieved the lowest error rates with an MAE of 0.1391 and an RMSE of 0.1612. These findings demonstrate that deep learning and ensemble methods significantly improve SCM decision-making in terms of accuracy, efficiency, and predictive power.