Supply Chain Delay Mitigation via Supplier Risk Index Assessment and Reinforcement Learning

Abstract

Supply chains are vulnerable to unforeseen delays, which might adversely affect delivery performance. Quantifying the risk profiles of each supplier based on their historic delivery patterns and forecast deviations can help make superior decisions in multi-supplier scenarios. This problem has been previously approached from linear programming and qualitative assessment perspectives; however, application of machine learning and reinforcement learning-based methods are still in a nascent stage. This paper proposes a machine learning technique to classify a supplier into one of four risk indices accurately on real-world datasets from Ericsson's supply hub. A reinforcement learning agent is also trained in a custom-modeled environment to split an order among multiple suppliers while minimizing the delays. Additionally, a working web-based tool is developed to demonstrate these techniques, that may be extended to other domains.