Answers are grounded in this paper and cite their sources.
Mutation testing is a powerful technique for ensuring software quality. However, the presence of the equivalent mutant’s problem is one of the crucial problems in mutation testing. In consequence of its existence, the effectiveness of mutation testing is underestimated, and a certain amount of useless overhead is produced. Many methods were applied for detecting and reducing the equivalent mutants. Although numerous equivalent mutant detection (EMD) methods have been proposed, such as traditional techniques, machine-learning techniques, and large language models (LLMs), they often face distinct challenges. In this paper, we focus on traditional technique challenges. Most traditional methods—such as those based on compiler optimization or symbolic execution—lack the systematic use of program graphs, such as abstract syntax trees (ASTs), control-flow graphs (CFGs), and data-flow graphs (DFGs), making it difficult to model key factors behind reachability, state infection, and propagation. To address this limitation, we propose the SMP_EM framework, a novel equivalent mutant detection approach based on program dependency graphs. The proposed approach performs static program analysis to trace the mutation propagation effect through the following steps: (1) localizing the mutation node, (2) constructing the CFG for the mutant, (3) extracting control-flow and data-flow dependencies based on the CFG, (4) integrating the CFG, control-flow, and data-flow dependencies to build the program dependency graph (PDG), and (5) performing mutation-guided propagation analysis, starting from the mutated node, to collect all mutation propagation paths (MPPs) and classify the mutant based on the effect of MPPs as either an equivalent mutant (EM), non-equivalent mutant (NEM), or inconclusive (UNKNOWN). To assess the efficiency of the proposed technique, it is evaluated on the MutantBech benchmark. The results compared with the labels in the benchmark show that our SMP_EM approach is more effective in detecting equivalent mutants, with an overall accuracy of 90.63%, a recall of 90.85%, a precision of 81.84%, an F1-score of 85.22%, and a Matthews correlation coefficient (MCC) score of 0.730. Our experimental comparisons demonstrate that significantly outperforms traditional compiler techniques (improving the F1-score by over 34.4% and recall by over 39.0% compared to TCE_Soot) and substantially surpasses standard machine-learning baselines (such as Random Forest, SVM, and Logistic Regression) by over 31.3% in recall and 12.9% in F1-score.
No takes yet. Share an insight, caveat, or question.
Marwa Hashem Shehata (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: