Abstract: Graph Neural Network (GNNs) has been widely employed in node classification over the past few years. However, existing research has predominantly focused on balanced datasets, whereas imbalanced data is prevalent. Traditional approaches to handling imbalanced datasets, such as resampling and reweighting, often require substantial preprocessing or proposing new network structures, which can introduce new biases and lead to information loss. An enhanced Bootstrap Aggregating (Bagging) ensemble learning method is proposed to address imbalanced graph datasets. It involves partitioning the data into k folds and training multiple distinct sub-models using GNNs as the base model. Finally, by fusing different models, the node classification accuracy is improved without introducing excessive preprocessing. Experimental results on imbalanced graph datasets demonstrate that the proposed method outperforms the base classifier in terms of accuracy and robustness. Additionally, it is observed that classification accuracy initially increases and then decreases with the increase of k.