GNN-Attention Framework for Efficient Test Path Coverage in Software Testing Using Control Flow Graphs

Abstract

Traditional test-generating methodologies are unsuccessful in modern software engineering because of path 
explosion, redundancy, and inadequate failure detection, which are caused by the dynamic behaviour and 
complexity of systems. In order to optimise test pathways from Control Flow Graphs (CFGs), this study offers a 
novel deep learning strategy that combines attention processes with Graph Neural Networks (GNNs). More fault 
discovery with fewer test cases is made possible by the output GNN-Attention model's dynamic emphasis on 
semantically important nodes and execution routes. Contextual embeddings are used to ascertain path relevance, 
source code is converted to CFGs, and datasets are used to describe structural and semantic information. High
relevance pathways are converted into executable test cases using test stubs or symbolic execution. Over 
coverage and classification accuracy, the model gains knowledge based on an aggregate losses function. The 
experimental result shows enhanced performance on 91.76% fault detection rate, 90.83% test coverage, and 
91.2% precision, along with increased scalability and efficiency. In comparison to conventional techniques, the 
model also shows a 9.3% improvement in defect identification with reduced test generation time and low 
computing cost. The rapidity, flexibility, and adoption of the model into testing CI/CD pipelines are validated by 
these results. The GNN provides a sophisticated, flexible, and context-aware method for creating software test 
routes, which helps to overcome the disadvantages of heuristic techniques and enhance software security in a 
range of contexts.