Bin Packing through Machine Learning

In this thesis project we propose a wide range of Machine Learning techniques for dealing with the Bin Packing problem. The business domain is transportation optimization, a popular application field of Operational Research methods. The work is inspired by a real project by the consulting firm Horsa Group. The aim is to inspect the business problem from a mathematical point of view and to focus on different state-of-the-art techniques involving Machine Learning. The objective is to give an overview of the different possible approaches for further developments and compare the pros and cons of possible solutions. We will also compare the performances of those techniques on generated example data and real-world data. The final goal is to reduce the costs of the shipping process by increasing efficiency. The focus will be on how the shipping pallets are composed, packing the items with an efficient and scalable framework. The road map consists in defining in a formal way the Operational Research problem and the business problem, to compare classical approaches with some of the methods that nowadays are more and more popular and involve Machine Learning techniques. Some of those approaches involve Deep Reinforcement Learning and Graph Neural Networks. Finally, we will inspect a wide range of possibilities for making the bin packing process more efficient, simulating different real case scenarios. The aim is to give a clear overview of future developments in Bin Packing Optimization algorithms. Those developments can make the company’s shipping software scalable and well-performing, with more efficient use of resources.