Minimizing Volumetric Drift: A K-Means Clustering Approach to Optimizing Corrugated Box Selection in E-Commerce Fulfillment

In modern logistics, "Volumetric Drift"—the discrepancy between product volume and shipping container volume constitutes a primary source of economic inefficiency and environmental waste. With the widespread adoption of dimensional (DIM) weight pricing by major carriers, the financial penalty for "shipping air" has escalated, yet Small and Medium Enterprises (SMEs) continue to rely on heuristic, static packaging inventories that fail to match shipment geometry. This paper proposes a data-driven framework for packaging optimization utilizing Unsupervised Machine Learning, Data set from AWS Open Data.

By applying K-Means Clustering to a dataset of 50,000 historical shipping logs (Length x Width x Height) from a mid-sized fulfillment center, we identify the geometric centroids of product clusters to derive an optimal "Golden Set" of box dimensions. Simulation results indicate that replacing standard legacy box sizes with four algorithmically generated clusters reduces average void space by 34% and total shipping costs by 12%. These findings demonstrate that low-compute, edge-deployable AI models can successfully mitigate volumetric inefficiency, offering a scalable solution for decarbonizing the "Last Mile" while enhancing profitability for logistics providers.