Predicting Hidden Links in Supply Networks

Abstract

Manufacturing companies often lack visibility of the procurement interdependencies between the suppliers within their supply network. However, knowledge of these interdependencies is useful to plan for potential operational disruptions. In this paper, we develop the Supply Network Link Predictor (SNLP) method to infer supplier interdependencies using the manufacturer's incomplete knowledge of the network. SNLP uses topological data to extract relational features from the known network to train a classifier for predicting potential links. Using a test case from the automotive industry, four features are extracted: (i) number of existing supplier links, (ii) overlaps between supplier product portfolios, (iii) product outsourcing associations, and (iv) likelihood of buyers purchasing from two suppliers together. Naïve Bayes and Logistic Regression are then employed to predict whether these features can help predict interdependencies between two suppliers. Our results show that these features can indeed be used to predict interdependencies in the network and that predictive accuracy is maximised by (i) and (iii). The findings give rise to the exciting possibility of using data analytics for improving supply chain visibility. We then proceed to discuss to what extent such approaches can be adopted and their limitations, highlighting next steps for future work in this area.