Comparing Transformer-Based Log Anomaly Detection With LLM Explainability In CI/CD Pipelines

With the DevOps setups of today, automating Continuous Integration and Deployment (CI/CD) pipelines is a necessity for achieving rapid software delivery. These pipelines generate enormous logs, which, despite being useful for debugging and monitoring, become cumbersome to analyze manually. In this c...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:2025 Intelligent Methods, Systems, and Applications​ (IMSA) S. 370 - 375
Hauptverfasser: Al-Aqrabi, Qasem Magdi, Al-Aqrabi, Hatem Magdi, Solayman, Marwa M.
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: IEEE 12.07.2025
Schlagworte:
Anomaly detection
Calibration
CI/CD pipelines
Context modeling
DevOps
DevOps automation
Explainable AI
Feedback learning
Pipelines
Semantics
Software
Training
Transformer models
Transformers
Visualization
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:With the DevOps setups of today, automating Continuous Integration and Deployment (CI/CD) pipelines is a necessity for achieving rapid software delivery. These pipelines generate enormous logs, which, despite being useful for debugging and monitoring, become cumbersome to analyze manually. In this comparative study, comparison of anomaly detection models on CI/CD pipeline logs, including classical, deep learning, and transformer-based approaches such as LogBERT is performed. An enhanced LogBERT-based classifier i s p resented, incorporating focal loss, grouped anomaly detection, and temperature scaling for better calibration and comparison with other Log anomaly detection models. The evaluation framework includes a feedback loop that allows user-labeled predictions to be reused for iterative retraining.
DOI:10.1109/IMSA65733.2025.11167168