Investigating the dynamics of spreading processes in real-world applications such as pathogen spread prediction, marketing, political events, etc has attracted the attention of researchers from a variety of fields. Influence-based information diffusion is one convincing attempt to solve the information diffusion problem. In this regard, most of the attempts suffer from certain drawbacks such as complexity, dependency on the underlying diffusion model, or low prediction accuracy. We have looked at this problem from a fresh perspective and come up with an innovative solution for solving it. Our hybrid approach falls at the intersection of three research areas: feature selection, graph embedding, and information dissemination. To discover the influential nodes in a network, we develop a method comparable to wrapper methods in feature selection, in which we employ the strength of graph convolutional neural networks (GCNs). The results of our implementation in Python on five datasets Cora, Email, Hamster, Router, and CEnew, under the susceptible–infected–recovered (SIR) model, approved that GCNFusion exceptionally outperforms benchmark methods by respectively around 3%, 5%, 5%, 2%, and 3%. Furthermore, the proposed method is a decent suit for real-world applications on complex networks due to its low computational complexity.