Explainable Artificial Intelligence Techniques for Software Development Lifecycle: A Phase-specific Survey

The "black-box problem" significantly limits AI adoption: Complex AI models that cannot explain their decisions restrict trust and widespread adoption of AI systems, making explainability a critical barrier to responsible AI deployment.

XAI research is heavily concentrated in software maintenance: The survey found that 68% of XAI research in Software Engineering focuses on the maintenance phase, while only 8% addresses software management and requirements—revealing significant gaps in applying explainability techniques across the full software development lifecycle.

Multiple XAI techniques exist but need broader application: The paper identifies seven key explainability methods (LIME, SHAP, concept-based explanations, rule extraction, attention mechanisms, counterfactual explanations, and example-based explanations) that can be applied throughout the development process, but their use remains unevenly distributed across different phases.

Explainability should be integrated from the start: By surveying XAI applications across all SDLC phases—from requirements elicitation through evolution—the paper emphasizes that transparency and interpretability need to be considered throughout the entire software development process, not just added at the end.

Stakeholders need clear AI explanations to verify outcomes: XAI techniques enable stakeholders to understand, trust, and verify AI-based decisions, which is essential for responsible AI systems that must be accountable to users and decision-makers.