Machine Learning-Enhanced CI/CD Pipelines in Kubernetes Environments: An Empirical Study

Abstract

Sachin Shivaji Raut* and Siddharth Shivaji Raut

The increasing adoption of cloud-native architectures and Kubernetes for software deployment presents various complexities for maintaining robust and efficient continuous integration and continuous deployment (CI/CD) pipelines. While machine learning (ML) holds promise for enhancing these processes, empirical investigation into its practical application and observed outcomes in real-world settings remains an area of active inquiry. This mixed-methods study empirically examines the implementation strategies and perceived efficacy of ML-enhanced CI/CD frameworks within Kubernetes-based environments. Data were col- lected through surveys and interviews involving 127 DevOps engineers, site reliability engineers, and cloud architects from twelve mid- to large-scale Software-as-a-Service organizations. Preliminary findings indicate that organizations integrating ML into their CI/CD pipelines reportedly observed an approximate 34% reduction in deployment failure rates and a 42% im- provement in mean time to recovery when compared to self-reported traditional approaches. These results may suggest potential advantages of ML integration in terms of operational resilience and efficiency. However, a deeper understanding of contextual factors and long-term implications is warranted, and these findings should be interpreted cautiously, considering the scope and specific characteristics of the studied organizations.

HTML PDF