FROM REACTIVE TO PREDICTIVE: HARNESSING KEDA AND CUSTOM MACHINE LEARNING FOR DYNAMIC POD MANAGEMENT

This paper presents a framework for enhancing Kubernetes autoscaling by integrating KEDA with a custom machine-learning algorithm. Unlike traditional autoscaling methods that rely on static thresholds, this approach leverages predictive analytics to dynamically adjust the number of pods based on anticipated workloads. The proposed system aims to optimize resource utilization, reduce costs, and improve application performance through intelligent scaling decisions. We discuss the algorithm design, hypothetical scenarios, and the expected benefits and challenges of this approach.