Computational Approaches to Packaging Optimization: Maximizing Load Efficiency in Global Supply Chain Logistics

Effective packaging plays a direct role in determining the costs of transportation and the sustainability of global logistics systems, where packages with different dimensions lead to inefficient loading due to volumetric loss, and existing strategies not usually optimize the loading process dynamically based on container size. Proposed research aims to improve loading by using an innovative intelligent computational system called the Modified Galactic Swarm-Based Layer-tuned Convolutional Attention Network (MGS-LCAN) to predict package suitability. Data is collected from the Kaggle logistics dataset, which contains 15,000 records with 21 input features such as container specifications, package characteristics, handling shipment characteristics, packaging indicators and 1 target distribution of package suitability. Preprocessing involves Missing Value Median Imputation for dealing with missing values in the input Data. In feature extraction, the technique used is Kernel Principal Component Analysis (KPCA), which aims to determine important volume and spatial features that affect the packaging configuration. The resulting features are then passed to the developed MGS-LCAN model, where MGS optimizes the convolutional layer parameters via an improved exploration-exploitation trade-off to accelerate convergence, and the optimized convolutional layer analyzes spatial arrangements inside the container. The architecture considers packaging optimization to reduce volume, with high, low medium suitability classes for global supply chains via adaptation to different geometric packages and container types. The Python implementation attains more efficient loading and better suitability than traditional packing heuristics, with an accuracy rate of 97.56% and F1-score of 97.55%.The MGS-LCAN framework provides a solution to improve packaging optimization in global logistics operations.