Graph neural network are a part of deep learning methods created to perform presumption on data described by graphs. Graph neural network is a neutral network that can straight away be applied to graphs. It provides a agreeable way for node level, edge level and graph level prediction tasks. Moreover, most GNN models do not account for long distance relationships in graphs and instead simply aggregate data from short distances (e.g., 1-hop neighbours) in each round. In this paper work, we carry out node classification using graphs which can be put into large graphs comprise of labelled and unlabelled nodes. Here we can predict the node embeddings of the unlabelled node by using an approach called message passing. For executingthis, we took Cora dataset, provided a overview of it, builded ourgraph neural network, splitted the data to test and train data, trained it and finally visualised the output. I