Multiple Fault Localization of statements with an evolutionary algorithm that combines Ochiai and LDA as fitness function
Romain Vacheret, Francisca Pérez, Tewfik Ziadi · Information and Software Technology · 2026
Context: Many Fault Localization (FL) approaches are designed to locate a single fault, but multiple faults typically exist in large programs. Multi-fault localization approaches at the statement level usually rely on Spectrum-Based Fault Localization (SBFL) without considering Information Retrieval-Based Fault Localization (IRFL). Objective: In this paper, we propose , a multi-fault localization approach that leverages an evolutionary algorithm to locate the code fragment that may be the cause of failure. The evolutionary algorithm can help to efficiently explore the sheer number of candidate code fragments, whereas the fitness function of the evolutionary algorithm combines Ochiai (SBFL) and LDA (IRFL) to assess the suspiciousness of each candidate code fragment. Method: We evaluate using the Defects4J benchmark, and we compare its results against four baselines: SBFL (Ochiai), IRFL (LDA and Blues), and hybrid (Ochiai and Blues). The comparison involves a statistical analysis to provide quantitative evidence of the impact of the results. Results: The results show that significantly outperforms the baselines by a large effect size. Specifically, our approach outperforms the best performing baseline by 51.7%, 19.6%, 11.5% and 11.5% for hit@1, 3, 5 and 10, respectively. Conclusion: Our work provides a new direction for the multi-fault localization of statements using an evolutionary algorithm and combining Ochiai and LDA as fitness function, which is valuable for the automated program repair community since FL directly impacts the accuracy of software fixes.