Enhancing Machine Learning Interpretability for Supply Chain Risk Management: A LLM-Powered Approach

Supply chain risk management increasingly leverages Machine Learning (ML) methods to predict and mitigate risks such as delays, disruptions, and other uncertainties. Despite recent advances, many state-of-the-art ML models are characterized as “black-box” systems, lacking interpretability and explainability that is essential for non-technical supply chain practitioners. This lack of transparency often leads to limited trust and reluctance among supply chain managers to adopt ML technologies, which hinders the effectiveness of risk mitigation efforts. To address these challenges, this paper introduces a novel application of Large Language Models (LLMs) to enhance the interpretability of ML-driven supply chain risk assessments. Specifically, we propose an LLM-powered framework that provides clear, accessible interpretations of model predictions, focusing on delay forecasting within supply chains. By translating technical insights into practical decision-making information, this approach aims to bridge the gap between complex ML outputs and stakeholder understanding. By harnessing the capabilities of LLMs for interpretability, this chapter contributes to a more transparent, reliable, and effective supply chain risk management process.