Interpretable Machine Learning to Improve Supply Chain Resilience, An Industry 4.0 Recipe

The growth of disruption risks, that is, risks with very low probability of occurrence and very high adverse impacts (e.g., pandemics, earthquakes, etc.), across the global supply chains, has increased over the past decades. In Industry 4.0 environment, the increasing applications of Artificial Intelligence (AI) throughout supply chain practices has led to the emergence of faster and more reliable decision-making methods when large volumes of data challenge the traditional methods. While applying machine learning (ML) techniques is well-documented in the supply chain risk literature, few studies focus on the interpretability of the outcomes achieved by ML. The present study aims to take a step towards fulfilling this gap by using machine learning algorithms on real-world data from an automotive supply chain In so doing, the performance data of 10 suppliers over two consecutive years were used. A clustering algorithm was first used to generate the labels based on the concept of resilience capacities. Then, since the interpretability of the results were a priority, two interpretable ML algorithms, Naive Bayes and decision tree, were chosen to classify the suppliers based on their performance with respect to each capacity. The results showed that for interpretable algorithms, decision tree could be potentially a better performing algorithm, yet Naive Bayes could provide more flexibility and insights through nomograms. Copyright (C) 2022 The Authors.