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

Abstract

Cloud-native architectures demand dynamic scaling mechanisms to balance performance, cost, and resource efficiency. Traditional reactive scaling methods often fail to address volatile workloads, leading to over-provisioning or service degradation. This paper proposes an AI-driven predictive scaling framework leveraging time-series forecasting, reinforcement learning, and hybrid models to anticipate resource demands and optimize cloud-native systems. We present a systematic evaluation of algorithms like LSTM and Prophet, integrated with Kubernetes orchestration, to demonstrate 35–40% cost reduction while maintaining 99.9% QoS compliance. Challenges such as data noise, model explainability, and ethical implications are critically analysed, alongside future directions in federated learning and energy-aware scaling.