Machine unlearning in supply chains

Supply chains are dynamic systems with constantly changing data, necessitating adaptive machine learning models. While prior research emphasizes integrating new data to enhance decision-making, the need to remove obsolete or harmful data from models remains underexplored. This paper addresses the challenge of efficiently removing undesired data, such as bankrupt suppliers or data errors, from already trained machine learning models via machine unlearning. We highlight supply chain challenges that can be addressed by machine unlearning, introduce a new unlearning method for practitioners that works without any parameter tuning, and demonstrate its effectiveness in an error unlearning case study.