Machine Learning for Sequential Decisions in Logistics
Fabian Akkerman is a PhD student in the Department of Industrial Engineering & Business Information Systems. (Co)Promotors are prof.dr.ir. M.R.K. Mes and prof.dr. M.E. Iacob†from the Faculty of Behavioural, Management and Social Sciences and dr. W. van Jaarsveld from Eindhoven University.
Logistics decision-making is becoming increasingly complex by uncertainty, real-time demands, and disruptions. Traditional operations research (OR) methods offer structured solutions but often struggle with dynamic and high-dimensional environments. Machine learning (ML) provides an alternative or complementary approach by leveraging data-driven techniques to adapt to changing conditions and optimize sequential decisions. This research investigates how ML can support logistics decision-making across supply logistics, distribution logistics, and revenue management.
In supply logistics, ML-based models are explored to improve inventory management under uncertainty. Decisions related to replenishment, sourcing, and inventory control are influenced by demand fluctuations and inaccuracies in inventory records. By integrating ML with optimization techniques, this research examines how hybrid approaches can enhance decision-making in these settings.
In distribution logistics, ML contributes to dynamic routing and customer selection. Predictive models estimate transportation costs to inform customer selection strategies, while reinforcement learning is used to optimize vehicle repositioning in response to real-time demand. These approaches aim to improve decision-making in large-scale and uncertain routing environments.
In revenue management, ML-driven models dynamically adjust the pricing and availability of logistics services, particularly in e-commerce. Strategies for offering out-of-home delivery options (e.g., parcel lockers) and optimizing attended home delivery time slots are examined. ML methods are used to adapt pricing and delivery offerings based on customer demand patterns, balancing efficiency and service quality.
This dissertation differentiates between data analytics, where ML extracts insights from historical data to support decision-making, and decision analytics, where ML directly optimizes sequential decisions in dynamic environments. A structured framework is proposed to guide the integration of ML models.
The objective of this research is to explore how ML can enhance decision-making in logistics under uncertainty. It contributes to the development of adaptive, data-driven frameworks that improve efficiency, flexibility, and resilience in logistics operations.
