In node classification tasks, traditional methods like LPA-GCN struggle with scalability and sensitivity to label noise. We propose LPA-GraphSAGE, combining the Label Propagation Algorithm with GraphSAGE’s sampling and aggregation techniques to address these challenges. Our model outperforms LPA-GCN in accuracy and efficiency, reducing computational costs. Its inductive capabilities also enable effective handling of dynamic graphs with new nodes. Experiments on benchmark datasets confirm that LPA-GraphSAGE is a robust and scalable alternative for node classification in complex graphs.