Long-standing challenges in the identification of risks in green supply chain finance stem from the complexity of data structures and the intertwining of non-linear information, leading to uncertainty. This study aims to construct a two-stage deep learning model that integrates Generative Adversarial Networks with a Residual Autoencoder (GAN-SAE) and Deep Neural Networks with Multi-Kernel Support Vector Machines (DNN-SVM) to achieve the intelligent and precise identification of risks. Based on a sample of 168 firms observed over the period 2015-2024, this study first employs a Conditional Generative Adversarial Network combined with a Residual Autoencoder (GAN-SAE) to generate and reconstruct high-risk samples. This process alleviates the scarcity of high-risk observations and mitigates feature redundancy, while simultaneously extracting discriminative risk representations. Building on this enhanced feature space, a hybrid classification framework integrating Deep Neural Networks and a Multi-Kernel Support Vector Machine (DNN-SVM) is further introduced to strengthen the mapping capability for complex nonlinear risk patterns and to achieve robust adaptive classification. Through this two-stage design, the proposed framework substantially improves the overall accuracy and stability of risk classification. Empirical results indicate that the proposed model demonstrates a significant advantage on the test set, achieving a classification accuracy of 0.923 ± 0.009, a recall rate of 0.893 ± 0.010, and an AUC value of 0.938 ± 0.008, outperforming the other comparative models. Moreover, the distribution of asset returns and liabilities of the generated samples aligns with that of the actual high-risk samples, validating the economic rationality of the model during the data augmentation phase. The risk heatmap results indicate that financial health and credit scores exhibit the highest contribution levels in the model's risk prediction, serving as the most sensitive feature dimensions for risk discrimination. In contrast, ESG performance and supply chain stability function as auxiliary variables that further refine the model's risk differentiation. The study achieves hierarchical expression of risk identification and optimization of feature coordination, offering a scalable intelligent recognition framework and methodological innovation for green financial risk control.
Xueping Gao (2026) studied this question.