Using Machine Learning for Analyzing Performance Metrics in Kubernetes

This article examines modern approaches to analyzing the performance metrics of containerizedapplications in Kubernetes using machine learning methods. The key classes of algorithms (regressionmodels, classification and anomaly detection methods, reinforcement learning) are analyzed, along withtheir advantages and limitations. The architectural principles of integrating machine learning into theKubernetes environment are presented, and typical application scenarios are proposed, including loadforecasting, automatic scaling, and failure and anomaly detection. A review of existing frameworks(Kubeflow, Katib, etc.) and monitoring tools (Prometheus, Grafana) is conducted, highlighting their rolein facilitating the implementation of MLOps pipelines within Kubernetes. Based on the findings,recommendations are formulated for selecting and combining machine learning methods depending onworkload profiles. Additionally, future research directions are outlined, including deeper integration ofreinforcement learning and predictive models to enhance the resilience of microservice architectures. Thisstudy is relevant for researchers and practitioners at the intersection of distributed computing, artificialintelligence, and DevOps who seek to integrate machine learning for automated monitoring, diagnostics,and performance optimization of Kubernetes clusters in scalable cloud infrastructures.