AI-Driven Predictive Scaling for Performance Optimization in Cloud Native Architectures

Beschreibung

This paper introduces an AI-driven predictive scaling framework designed for cloud-native architectures, addressing the limitations of traditional reactive methods in handling dynamic workloads. By leveraging time-series forecasting techniques (LSTM, Prophet) and reinforcement learning, the framework integrates with Kubernetes to optimize resource allocation. The study demonstrates a 35–40% reduction in operational costs while ensuring 99.9% Quality of Service (QoS). It also explores challenges related to data noise, model transparency, and ethical considerations, and outlines future advancements in federated learning and energy-efficient scaling strategies.