<p>Enterprise credit assessment plays a pivotal role in ensuring the stability of financial systems by identifying potential credit risks and facilitating efficient capital allocation. However, the coexistence of multiple high dimensional features, dynamically evolving relational structures, and imbalanced data distributions often leads to inconsistencies between feature representation and decision boundaries, resulting in representational bias and degraded predictive performance. To address these challenges, this paper proposes a Confidence Driven Dual Domain Fusion Adaptive Graph Neural Network (DCSGN) for enterprise credit assessment. Specifically, a dual domain feature fusion module is designed to integrate local aggregation and channel mixing information, thereby enhancing the model’s capability to capture both local correlations and cross dimensional dependencies among enterprises. An adaptive gated graph is then constructed, where learnable gating parameters dynamically adjust edge weights to highlight critical relationships while suppressing redundant connections. Furthermore, a confidence driven sample reweighting mechanism is developed to adaptively allocate training weights based on prediction confidence, effectively mitigating evaluation bias caused by data imbalance. Experimental results on two benchmark datasets demonstrate that DCSGN consistently outperforms several state of the art methods in terms of accuracy, F1 score, and AUC, confirming its effectiveness and robustness for enterprise credit assessment tasks.</p>