Machine learning enhanced closed loop supply chain control

Abstract

Closed-loop supply chains use both new components and products returned by customers, its planning means considering many possible futures at once, but an optimization model cannot carry thousands of scenarios. This thesis asks how clustering, a machine learning method, can reduce a large scenario set into a small one that is still good enough to control the chain. The choice of the clustering method is essential for good results and representatives, because, first it saves computational time, and help the decion maker choose the plan to apply. The clustering shows which uncertain parameters change the decisions and which ones the chain absorbs without changing its way of operating.