AutoScaleAI: An AI-Powered Framework for Cloud Resource Optimization Using Telemetry Analytics

Read full text Summarise with VIRAT
Authors: Faiz Gouri, Chakra Pavan Kumar Kota, Mahesh Shivnatri, Kuber Jain
Cited by: 0
References: 27

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Cloud infrastructure operators face increasing challenges in balancing performance, cost, and reliability due to dynamicworkloads and complex interdependency. This paper presentsAutoScaleAI, an AI-driven telemetry analytics framework designed to proactively optimize cloud resource utilization and reduce operational costs.The framework integrates a hybrid prediction model Gradient Boosting Machines (GBM) and Long Short-TermMemory (LSTM) networks, used to forecast short-horizon CPU, memory, and network utilization trends, enabling proactive scaling decisions. It further incorporates a two-tier anomaly detection engine Isolation Forest (IF) and Variational Autoencoders (VAE) to identify sudden workload surges, latency deviations, and telemetry drift that may violate service level objectives. Telemetry data modeled on real-world Azure deployments (synthetic simulations validated against historical traces) are used to train and evaluate the system across compute, storage, and network dimensions. In addition, we conduct a controlled real-world pilot on Azure Kubernetes Service (AKS) with live telemetry and production APIs, validating end-to-end feasibility and quantifying operational overhead. Experimental results show that AutoScaleAI achieves up to 38% cost reduction and a 35% increase in resource efficiency, outperforming traditional threshold-based and deep reinforcement learning baselines. Additionally, the system achieves a decision latency of 235 milliseconds, making it suitable for real-time cloud optimization scenarios. AutoScaleAI is further validated through real-world case studies in e-commerce, finance, and healthcare environments. This work contributes a scalable and adaptive solution for AI-powered cloud management, offering practical benefits for both hyperscale providers and enterprise consumers.

Fields of Study & Subjects

Cloud Computing and Resource Management Software System Performance and Reliability Energy Efficiency in Computing

Publication details

Journal / source
IEEE Access
Volume
14
Pages
49082–49098
Publication date
1 January 2026
Language
EN