Evaluating Supply Chain Network Designs: An Approach Based on SNA Metrics and Random Forest Feature Selection

Abstract

Supply chain network design is an important decision-making problem affecting the long-term profitability of firms. Evaluating the performance of supply chain network designs can help decision-makers to select the network configuration that meets the business specifications while operating at a reasonable cost. In this study, Social Network Analysis (SNA) metrics are used to evaluate the performance of closed-loop supply chain (CLSC) Network designs in terms of resilience when exposed to disruptions and the balance of flows. CLSC Networks accommodate the flow of returned products from the customers for recycling, remanufacturing, or disposal, increasing the design complexity compared to traditional supply chain networks. The proposed approach involves custom-designed network-level SNA metrics and random forest (RF) feature selection which are computationally low-cost approaches. The proposed metrics are implemented in an R package titled NetworkSNA and shared on GitHub, and RF feature selection method is performed in python. The optimal and near-optimal network designs from a CLSC Network based on real data are used as a case study. The metric values are interpreted into practical recommendations to compare the alternative CLSC Networks.