EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

Reducing AutoML Pipeline Failures in Heterogeneous Cloud Environments via Retrieval-Augmented Fault Prediction

Zihan Liu, Ruihan Ma

International Journal of Advanced Engineering and Technology Research · 05.06.2026 · 2 (2) · 38-45

DOI: 10.54097/ath09617 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Advanced Engineering and Technology Research, online erschienen am 05.06.2026. Verfasst von Zihan Liu und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Automated Machine Learning (AutoML) pipelines deployed in heterogeneous cloud environments are increasingly susceptible to runtime failures caused by resource contention, hardware diversity, and dynamic workload fluctuations. These failures impose substantial operational overhead and compromise the reliability of large-scale machine learning workflows. This paper introduces a Retrieval-Augmented Fault Prediction (RAFP) framework designed to anticipate AutoML pipeline failures by coupling a dense vector retrieval module over a structured historical failure knowledge base with a gradient-boosted fault classifier. The retrieval module encodes contextually similar past failure records as auxiliary features, enabling the prediction model to condition its output on domain-specific failure patterns rather than relying solely on real-time telemetry signals. The RAFP framework is evaluated against four baseline systems — logistic regression (LR), random forest (RF), gradient boosting (GB), and long short-term memory (LSTM) networks — using the Google Cluster Trace v3 and the Alibaba Cluster Trace 2018. Experimental results demonstrate that RAFP achieves a macro-averaged F1-score of 0.891 and reduces pipeline disruption events by 37.4% relative to the strongest baseline. Ablation studies confirm that the retrieval component yields consistent performance improvements across heterogeneous node configurations. These findings indicate that retrieval-augmented reasoning represents a practically effective complement to existing proactive fault prediction architectures for cloud-hosted AutoML systems.

Zitieren

APA

Liu, Z., & Ma, R. (2026). Reducing AutoML Pipeline Failures in Heterogeneous Cloud Environments via Retrieval-Augmented Fault Prediction. International Journal of Advanced Engineering and Technology Research, 2(2), 38-45. https://doi.org/10.54097/ath09617

Vancouver

Liu Z, Ma R. Reducing AutoML Pipeline Failures in Heterogeneous Cloud Environments via Retrieval-Augmented Fault Prediction. International Journal of Advanced Engineering and Technology Research. 2026;2(2):38-45. doi:10.54097/ath09617

BibTeX
@article{Liu2026,
  author = {Liu, Zihan and Ma, Ruihan},
  title = {Reducing AutoML Pipeline Failures in Heterogeneous Cloud Environments via Retrieval-Augmented Fault Prediction},
  journal = {International Journal of Advanced Engineering and Technology Research},
  year = {2026},
  volume = {2},
  number = {2},
  pages = {38--45},
  doi = {10.54097/ath09617}
}
Kostenlos starten →