A Three-Stage Stochastic Facility Location Problem: Integrating Design of Experiments and Artificial Neural Networks for Optimized Decision Making

This paper presents a novel approach to solving the stochastic facility location problem with uncertain customer demand, using a combination of Design of Experiments and Artificial Neural Networks. The study aims to optimize the facility location decision under two distinct cases of demand variability: small and large fluctuations in customer demand. In contrast to traditional deterministic models, this research incorporates demand uncertainty by focusing on four key customers whose demand is highly variable, thereby influencing the facility location decisions. The proposed methodology involves modeling the relationship between input variables (customer demand) and output variables (total cost and site selection) using a second-order polynomial model within the neural network framework. The approach is validated through a set of experimental results, with a comparison between predicted and actual costs showing a deviation of less than 3%. This demonstrates the model's robustness and reliability in making facility location decisions. The results indicate that factors DC10, DC12, and interactions between customer demands (e.g., DC1DC5 and DC10DC12) are significant in determining cost-effective facility locations. Furthermore, the study contributes to the optimization of distribution networks under uncertainty, offering a practical and effective tool for decision-makers in logistics and supply chain management. The proposed model is particularly useful in contexts where demand fluctuations are substantial, providing a reliable and computationally efficient solution to the facility location problem.