AI-Driven Cloud Administration: A Literature Review and Comparative Synthesis of Forecasting, Resource Allocation, Cost Optimization and Load Balancing Approaches

Abstract

This review article examined AI-driven approaches for cloud administration through a structured literature review and comparative synthesis of studies published between 2016 and 2025 (N = 57). The review focused on four interdependent administrative functions: predictive workload analysis, dynamic resource allocation and scheduling, cost–energy–QoS optimization, and AI-enhanced load balancing under reliability and security constraints. Publications were retrieved from major scholarly databases and were screened using eligibility criteria requiring direct relevance to cloud operations, explicit use of AI/ML/optimization for operational decision-making, and reported operational metrics or comparative evidence. The synthesis indicated that short-horizon forecasting models generally reduced over-provisioning and supported proactive scaling, but forecasting was often evaluated in isolation, limiting end-to-end evidence for sustained SLO improvement under concept drift and multi-cloud variability. Reinforcement learning and meta-heuristic schedulers commonly improved utilization and makespan relative to rule-based baselines, although many studies relied on simulator settings and reported limited reproducibility and generalization under realistic constraints. Cost- and energy-aware methods frequently reduced OPEX and energy via consolidation, DVFS, and cost-aware placement, yet they exposed trade-offs with QoS stability and used heterogeneous modelling assumptions. AI-based load balancing approaches improved latency and robustness in burst and failure scenarios, while explainability and portable trust/reliability metrics remained underdeveloped. Based on cross-stream evidence, a conceptual reference framework was derived that linked forecasting, scheduling, cost–energy objectives, and load balancing as a unified decision pipeline and highlighted gaps in benchmarks, portability, and transparency.