New Research: Machine Learning in Logistics Decision Making
Improving logistics processes through data analysis and smart technologies

In the fast-changing world of logistics, making the right decisions is crucial for the success of organisations. The use of machine learning (ML) offers new opportunities to optimise logistics processes, especially in situations characterised by uncertainty and time pressure. Fabian Akkerman earned his PhD degree Cum Laude with his dissertation entitled "Machine Learning for Sequential Decisions in Logistics". In his PhD track, he investigated how ML can contribute to better decision-making in three key logistics areas: supply logistics, distribution logistics and revenue management. His work is an important contribution to the DynaPlex project, funded by TKI Dinalog, and contributes to the further development of smart and sustainable logistics solutions.

Core components of the study

Supply Logistics: In the world of inventory management, the thesis uses a hybrid ML and optimisation model for replenishment and inspection decisions, combining neural networks with traditional operational research techniques. Furthermore, reinforcement learning (RL) is applied to dual sourcing, which increases supply chain resilience in the face of demand uncertainty. In addition, an innovative neighbourhood search algorithm is developed for large inventory management issues.
Distribution Logistics: ML is deployed to improve customer choice and routings. The research provides predictive models that help estimate transport costs for customer selection over multiple time periods. RL is also applied to dynamic vehicle routing problems, allowing real-time adjustments when demand fluctuates.
Revenue Management: In this domain, ML is used to dynamically adjust logistics services, for example through smart pricing and customer selection for parcel lockers and through decision-focused learning to optimise time slots for home delivery.

Data analysis versus Decision analysis

The research makes an important distinction between data analysis, which draws insights from historical data to support decisions, and decision analysis, which focuses on directly optimising sequential decisions. The thesis introduces a structured framework to effectively integrate ML into logistics processes, resulting in adaptive and data-driven decision-making processes that increase efficiency, flexibility and resilience.

Practical applications for the logistics industry

This research provides a solid foundation for implementing machine learning in logistics, from inventory management to dynamic distribution and revenue management. It can help companies make faster and more accurate decisions in an increasingly complex market.

Fabian Akkerman, who successfully defended his thesis at the University of Twente, is looking forward to further expanding his expertise as a postdoctoral researcher.