AI-Driven Root Cause Analysis for Cloud-Native Systems Using Logs, Metrics, and Distributed Traces

Authors

  • Madhava Reddy Muli Author

DOI:

https://doi.org/10.68306/tphr1c54

Abstract

Cloud-native systems pose big data challenges in root cause analysis because of distributed microservices, dissimilar telemetry, and interwove service relationships. The framework proposed by this research suggests the use of AI to help with systematically diagnosing incidents by combining logs, infrastructure metrics, traces distributed among devices deployed, and information on deployment. Experimental evaluation is aided by a synthetic data set of 6,000 observations and 120 incidents plus six categories of root causes. The stages of the data preprocessing, statistical analysis, anomaly detection, feature engineering at the incident level, and validation of the evidence are performed with Python. The comparative analysis of the four classification methods, that is, Majority Baseline, Logistic Regression, Random Forest, and Gradient Boosting, is examined. The results show a 100% accuracy and F1-score of both Logistic Regression and Random Forest and 96.67-percent accuracy of Gradient Boosting. Correlation analysis indicates that there is a strong correlation between request latency and error rates (r=0.77), and ablation experiments create the significance of the operational metrics. The framework also prepares organized evidence to further reasoning with the assistance of LLM, though the validation of production is needed to evaluate the feasible generalizability.

Downloads

Published

2025-07-22