A Two-Stage Deep Learning Model for Supply Chain Risk Evaluation with Multi-Source Data

The multi-source nature of supply chain data, temporal characteristics, and inherent risk class imbalance pose significant challenges to evaluate risks accurately. This study proposes a novel two-stage deep learning model for supply chain risk evaluation through multi-source data. In the first stage, the Conditional Generative Adversarial Network (CGAN) is employed to generate minority-class samples to address class imbalance. The second stage incorporates an improved BiLSTM-1DCNN model that simultaneously captures long-term dependencies and global temporal patterns while extracting local features. The proposed model integrates attention mechanism and residual connection to focus on critical features while mitigating gradient vanishing/explosion issues caused by increasing network complexity. Experimental results on real-world datasets demonstrate the model's superior performance, achieving 91.36% precision, 88.67% recall, and 88.93% F1-score, which outperforms the baseline model. This study provides an effective decision support tool for supply chain risk management in practice.