Low-carbon supply chain logistics risk prediction using meta-learning-based graph convolutional network on prototype...

Area of Science:

• Supply Chain Management
• Risk Analysis
• Machine Learning

Background:

• Predicting rare, high-impact risks in low-carbon supply chains (LCSCs) is crucial for sustainability.
• Data imbalance and complex interdependencies pose significant challenges to existing risk prediction models.

Purpose of the Study:

• To develop an advanced model for accurate and cost-effective early-warning of risks in LCSCs.
• To address data imbalance and complex interdependencies in LCSC risk prediction.

Main Methods:

• Proposed a Meta-Learning-based Graph Convolutional Network on Prototype Space (ML-GCNPS).
• Modeled LCSCs as graphs using multi-modal, firm-level carbon emissions and supply-chain data.
• Employed an adaptive Vertex-to-Edge (V2E) network for graph topology construction and risk propagation via GCN.

Main Results:

• ML-GCNPS significantly outperformed state-of-the-art baselines.
• Achieved an Area Under the Precision-Recall Curve (AUPRC) of 0.850.
• Reduced the False Negative Rate (FNR) to 0.080 with a Weighted Average Cost (WAC) of 45.

Conclusions:

• The ML-GCNPS model demonstrates practical value for accurate and cost-effective risk early-warning in LCSCs.
• The proposed method effectively handles data imbalance and complex interdependencies.
• Results confirm the model's capability for robust LCSC risk prediction.