Automated Root Cause Analysis in SRE: Applying Machine Learning to Incident Telemetry and Change Events

Abstract

Root cause analysis (RCA) remains time-consuming due to fragmented telemetry and complex change histories. This paper proposes a machine learning approach that correlates incident telemetry with deployment events, configuration changes, and dependency health signals to suggest likely fault domains. The study evaluates graph-based correlation, anomaly scoring, and explanation techniques that keep recommendations actionable for engineers. Results show reductions in triage time and improved first-guess accuracy during high-severity incidents.