.png){kind=icon}
.png){kind=icon}
Return
Graph neural network with generative adversarial training for node classification on class imbalanced data
.png){kind=icon}
PRE
AI
DOI：10.1016/j.engappai.2025.112264.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Node classification in class-imbalanced graph data remains a critical challenge, as traditional graph neural networks (GNNs) either assume class-balanced graph structures or inadequately address class imbalance. This often results in predictive bias, where majority classes are favored while minority classes are underrepresented. To overcome this limitation, this study introduces a novel graph neural network with generative adversarial training (GNN-GAN), where the GNN extracts latent features from input node attributes balanced through data synthesis using a conditional generative adversarial network (GAN) and data fusion strategy. The GNN and GAN are trained synchronously to ensure GAN synthesizes samples that match real data distribution, while GNN adjusts to the quality of synthesized data in a timely manner. A data fusion strategy combines synthetic and real samples to mitigate class imbalance and maintain classification accuracy. Experiments on several benchmark graph datasets demonstrate that the GNN-GAN consistently outperforms state-of-the-art baselines. A comprehensive ablative study further validates the advantages of the synchronized training procedure, offering insights into the model's robustness across graph datasets with varying structures and imbalance ratios.
Journal
IF:
8
Papers:
5.3K
Citations:
3.5W
Organization
No organization information available