SCR Analysis of End-to-End Explainability in Automl and Autoxai and Synergistic Justification of Architectural Integration

Automated Machine Learning (AutoML) has rapidly evolved in recent years and is increasingly being adopted in practical settings. However, the increasing level of automation has intensified the transparency problem, giving rise to the phenomenon of “double opacity” — that of both the model and the procedure used to select it. Explainable Artificial Intelligence (XAI) methods are predominantly applied post hoc and are not integrated into the AutoML optimization loop. This study presents a scoping review of research at the intersection of AutoML and XAI, aimed at systematizing existing integration approaches and identifying architectural limitations. The literature search covering 2015–2026 was conducted in Scopus, Web of Science, eLibrary, IEEE Xplore, ACM Digital Library, PubMed, and arXiv in accordance with PRISMA-ScR guidelines. The results show that the dominant paradigm follows a sequential scheme, “AutoML → model → XAI,” in which interpretability is not incorporated into the optimization criterion and does not influence the search trajectory. Key gaps are identified, including the lack of synergistic orchestration, the exclusion of explainability from the objective function, and the absence of a feedback mechanism. Architectural requirements for AutoXAI systems are formulated, including multi-objective optimization, bidirectional component connectivity, consideration of the computational cost of explanations, and ensuring decision traceability.