[2D6-GS-2] Machine learning: Bayesian estimation

[2D6-GS-2-05] Advances in Logistics Cost Reduction: Applying Bayesian Optimization to Cardboard Box Sizing

〇Atsuyoshi Muta¹, Shota Nishioka¹, Yasuyuki Mitsui¹, Kazuhiro Koike¹, Kota Fukamachi², Kiri Miura², Naoki Kobayashi², Kenji Tanaka² (1. ASKUL Corporation, 2. Univ. of Tokyo)

Keywords:Bayesian optimization , black-box optimization, e-commerce logistics, cardboard box size design, bin-packing problem

In e-commerce logistics, the design of cardboard box sizes directly impacts shipping costs. Determining more cost-effective box dimensions requires considering a vast number of combinations. While the packaging of orders as a 3D bin-packing problem has been studied, research on optimizing box size design itself as a mathematical problem has been scarce. This study proposes a method using Bayesian optimization to calculate box dimensions for minimizing shipping costs. The novelty of our approach lies in dividing the problem into two parts. The first is a combinatorial optimization problem aimed at selecting the smallest possible box that can pack all items of an order. Once the box size is determined, the shipping cost is fixed, leading to the second problem: minimizing costs for the set of orders using Bayesian optimization. The advantage of employing Bayesian optimization is its ability to prioritize candidates with a higher potential to improve the objective function, thus reducing the need to frequently solve the computationally intensive first problem. Simulations using real data showed a 0.55% cost improvement. Given the large base number, this yields significant benefits.