When forecasting is not enough: using simulation and optimisation to improve inventory decisions
Data science is often associated with forecasting what might happen next. But in many operational settings, prediction is only part of the problem. The bigger challenge is deciding what to do when demand is uncertain, objectives conflict and business conditions keep shifting.
A recent paper, Advancing EOQ models with stochastic demand and multi-objective optimisation through simulation-based search, explores this challenge through the lens of inventory management. Rather than treating demand as fixed and predictable, the paper develops a simulation-based framework that models demand as seasonal, variable and uncertain. This allows different replenishment decisions to be tested over time, giving a more realistic view of where shortages, excess stock and service failures may occur.
For data scientists, the paper is interesting because it moves beyond prediction and into prescriptive analytics. It asks how analytical models can support better decisions once uncertainty has been recognised. By simulating inventory behaviour across a full annual cycle, the framework creates a decision-testing environment where different ordering policies can be explored before they are applied in practice.
The model also uses multi-objective optimisation to reflect the trade-offs that organisations face in the real world. Inventory decisions rarely come down to one simple target. Businesses may want to reduce total cost, improve service levels and avoid holding unnecessary stock, but these aims can pull in different directions. The paper applies NSGA-II, a multi-objective optimisation method, to identify a set of compromise solutions rather than a single answer. This Pareto-based approach helps decision-makers understand the balance between competing priorities.
Another important data science angle is the use of surrogate-assisted modelling. Simulation can be computationally expensive, particularly when many candidate policies need to be tested. By incorporating a machine-learning surrogate, the framework aims to reduce the cost of search while still capturing the behaviour of a more complex operational system. This combination of simulation, optimisation and machine learning reflects a wider shift in data science, from building models that describe the world to building systems that help people act within it.
The paper sits within a long tradition of research that has extended classical EOQ models beyond their original assumptions. Its contribution lies in bringing together stochastic demand, multiple replenishment policy types, multi-objective optimisation and surrogate support within a single modelling framework.
For organisations managing stock, supply chains or purchasing decisions, the message is clear: effective data science is not only about forecasting demand accurately. It is also about testing decisions, understanding trade-offs and helping teams choose robust actions in uncertain environments.
References:
https://www.mdpi.com/1999-4893/19/5/415
https://www.sciencedirect.com/science/article/pii/S0898122108004884