What the study is dealing with is the presentation of a brand-new way of software defect prediction using control flow graphs connected to a convolutional neural network. The old-fashioned defect prediction techniques which rely on the static metrics and a tree structure, are comforted in the fact that they are neither able to describe semantics which are dynamic and thus, complex, thus, embedded in software programs. Our approach sets forth the control flow graphs (CFG) of the assembly code, which is the code that describes the path of the running program with high level of details. With the assistance of convolutional neural networks (CNN) on these graphs, our models can easily automatically realize the pattern of deficits in the software codes. The results from applying our model in four real-world datasets demonstrate a large success in predicted accuracy as well as a greater performance than the traditional and other deep learning (DL)-based approaches. Such enhancement shows the advantages of graph-based deep learning frameworks for the detection of software faults, after this entire enhancement leads to cost reduction for debugging and software reliability and performance improvement.