Workload Forecasting Techniques in Cloud Computing: A Review and Future Directions

Workload forecasting in Cloud Computing is the process of predicting future resource demands (CPU, Memory, Bandwidth and Disk I/O). It uses historical data to optimize resource utilization and management. Accurate prediction of workload leads to lower operational costs, preventing Service Level Agreement (SLA) violations and avoiding over- or under-provisioning. Researchers have developed multiple approaches to forecast workload, broadly categorized into Statistical methods, Machine Learning (ML) and Deep Learning (DL) techniques. This review primarily discusses the significant advancements in workload forecasting within Cloud Computing environments over the past few years. It mainly focuses on ML, DL and hybrid/ensemble methods used from 2020 to 2025. The present study identifies various issues through the analysis of existing research, like dealing with non-linear workloads, high-dimensional temporal dependencies and managing data variability. The review reveals that hybrid models have superior performance compared to standalone methods. It further proposes future research directions, including transfer learning applications, multi-resource prediction frameworks, energy-aware computing and uncertainty-aware forecasting models.