Deep Learning Optimization for Credit Risk Assessment in Supply Chain Finance Under the Game Theory Framework

Supply chain finance, as a core financial innovation model to alleviate the financing difficulties of small and medium-sized enterprises (SMEs) and enhance supply chain coordination efficiency, the accuracy of its credit risk assessment directly determines the safety and sustainability of financial services. Traditional assessment methods have shortcomings such as insufficient characterization of multi-agent game behaviors and weak ability to capture non-linear risk factors. This paper deeply integrates game theory and deep learning to construct an optimized model for credit risk assessment in supply chain finance. Firstly, based on Stackelberg game theory, it analyzes the strategic interaction mechanism among core enterprises, SMEs, and financial institutions, and derives the profit functions and game equilibrium conditions of each agent. Secondly, it designs an Attention-Based Bidirectional Long Short-Term Memory (Attention-BiLSTM) model, incorporating key strategic variables obtained from game analysis into the input feature system. Finally, empirical tests are conducted using 5,000 actual transaction data from a supply chain finance platform. The results show that the proposed model is significantly superior to comparison models such as logistic regression and single LSTM in core indicators including accuracy and recall rate, and game features rank among the top in contribution to risk assessment. This study provides interdisciplinary theoretical support and practical solutions for the precise control of credit risks in supply chain finance.