Explainable Artificial Intelligence in Software Engineering: Current Trends, Gaps, and Future Directions

Explainable Artificial Intelligence (XAI) has gained increasing attention for improving the transparency of artificial intelligence systems. However, there is a less clear understanding of its adoption and research patterns in the field of Software Engineering (SE).This systematic literature review focuses on the use of XAI within SE and the prospects of cross-domain learning. To achieve this, twenty-nine studies applying XAI in Software Engineering (SE), published between March 2020 and April 2024, were reviewed, and their findings were compared with XAI practices reported in healthcare and finance. The findings show that SE research mostly uses local explanation techniques: 86% of the studies employ instance-level methods, such as Local Interpretable Model-agnostic Explanations (LIME), whereas global explanation techniques that provide an overall view of the model are used less frequently. In terms of application domains, XAI studies in SE are mainly focused on software defect prediction, while other domains, such as software testing, and effort estimation remain comparatively underexplored. Regarding implementation tools,scikit-learnis the most frequently used machine learning (ML) library in XAI-based SE studies, and less often,TensorFlow, PyTorch, andRbased-frameworks are used. The current study is more comprehensive in terms of scope compared to the previous review, as it addresses four research questions, examines the types of explanations, and clearly mentions the utilization of ML libraries. The findings indicate that future SE studies should incorporate global explanation techniques, diversify XAI methods, and extend the research to underrepresented SE tasks. In addition, clear reporting of tools and libraries is essential to enhance reproducibility. Overall, although XAI in SE is progressing, the domain remains at an early stage, and insights from other fields may help improve interpretability, reliability, and trust in SE systems.