Disruption evaluation in end-to-end semiconductor supply chains via interpretable machine learning

Abstract

COVID-19 has posed unprecedented challenges to global health and the world economy. Two years into the pandemic, the widespread impact of COVID-19 continues to deepen, impacting different industries such as the automotive industry and its supply chain. This study presents a hybrid approach combining simulation modeling and tree-based supervised machine learning techniques to explore the implications of end-market demand disruptions. Specifically, we apply the concept of born-again tree ensembles, which are powerful and, at the same time, easily interpretable classifiers, to the case of the semiconductor industry. First, we show how to use born-again tree ensembles to explore data generated by a supply chain simulation model. To this end, we demonstrate the influence of varying behavioral and structural parameters and show the impact of their variation on specific key performance indicators, e.g., the inventory level. Finally, we leverage a counterfactual analysis to identify detailed managerial insights for semiconductor companies to mitigate adverse impacts on one echelon or the entire supply chain. Our hybrid approach provides a simulation model enhanced by a tree-based supervised machine learning model that companies can use to determine optimal measures for mitigating the adverse effects of end-market demand disruptions. We close the loop of our analysis by integrating the findings of the counterfactual analysis backward into the simulation model to understand the overall dynamics within the multi-echelon supply chain.