An AIGC Model for FinTech Risk Identification of Heterogeneous Participants from the Perspective of XAI

In the fintech ecosystem, banks, listed companies, and the stock market form a non-linear risk network through the "equity pledge - market value fluctuation - financial linkage" mechanism. Traditional models have hit a bottleneck due to fragmented data, insufficient characterization of transmission mechanisms, and the "black box" issue. To address the core problems of missing key cross-entity variables, difficulty in capturing non-linear transmission, and the challenge of balancing prediction and interpretation, this paper constructs the "XAI-AIGC-PXS" model: lightweight AIGC is used to generate 4 types of cross-entity risk features to fill data gaps; PSO-optimized XGBoost is employed to enhance prediction performance; and a "global-local" dual-level interpretation framework is built in combination with SHAP. Empirical results based on 11.6 million observation records from CSMAR show that Model D, which integrates dual feature sets, performs the best, with an accuracy of 72.6%, precision of 71.3%, recall of 61.4%, F1-score of 63.3%, and AUC of 0.788, significantly outperforming traditional models. The attribution results are consistent with the risk transmission logic and demonstrate good robustness. This research breaks through the "bank-enterprise" binary framework, constructs an intelligent risk control paradigm of "data enhancement - accurate prediction - transparent interpretation", expands the application boundary of related theories, and provides efficient technical tools for risk control of financial institutions, enterprise decision-making, and regulatory governance.