Recent years have seen a significant increase in the research work regarding graph-based datasets and GNN. Research in node classification in a graph, link prediction between nodes, anomaly detection in graphs, etc. have become significant in various applications. Graph neural networks are used to perform various tasks on graphs one such task is anomaly detection which is identifying anomalous nodes in the graph. This paper proposes a novel GNN-based architecture that uses GANs for identifying anomalies in the graph. The model outperforms other MLP, RNN-based architecture. GNN is used because a graph is a complex data structure difficult to represent in a latent shape due to variable shapes, etc. GNN helps us to create proper graph embedding which helps in incorporating maximum information from the graph. The model consists of a discriminator network, a generator network, and an encoder made up of graph convolutional layers as a base neural network. The proposed model especially focuses on a single graph-based dataset where the whole graph is passed as input and anomalies are identified based on the anomaly score calculated by combining different reconstruction losses calculated from the encoder and generator. A higher anomaly score is observed for the anomalous nodes. The model proposed outperforms all the traditional outlier detection models when compared based on ROC metrics.