To ensure the quality of software source code, numerous software testing approaches have been studied. Software is now integral to numerous devices, enterprise services, and public services. Although the demand for software quality has increased, Software Science has yet to provide a definitive solution for bug prediction methodologies. In this study, we propose a novel bug prediction methodology for software testing using Graph Neural Network (GNN) techniques. We attempt to apply the machine learning technique of Graph Convolutional Neural Networks (GCN) to Control Flow Graphs (CFG) generated from the tri-address information of the test target source code. In the CFG, multiple graph centrality values are utilized as graph feature for bug prediction. Hence, our bug prediction model based on graph neural network (BP-GNN) exhibits a better result with an accuracy value of 82%. This result represents an 15% improvement compared to the outcomes of previous study using Akaike Information Criterion (AIC) with graph centrality annotation for same CFG data.