Hybrid Ensemble Approach for Accurate Workload Prediction in Dynamic Cloud Environments

Time series forecasting is important for cloud computing because it helps keep services running, ensures that Quality of Service (QoS) is maintained at a high level even when workloads change, and makes sure that resources are used well. In the actual world, cloud workloads do not stay the same all the time. Rather, they evolve throughout time, exhibiting cyclical trends, nonlinear fluctuations, and periods of elevated demand. This complicates the utilization of conventional forecasting methodologies. This study presents a hybrid ensemble-based approach for short-term cloud workload prediction that enhances adaptability through workload-aware modeling to tackle this issue. The method employs feature extraction and K-means clustering, complemented by a dynamically weighted ensemble of Prophet, Long Short-Term Memory (LSTM), and Linear Regression models. NASA HTTP server logs and ClarkNet WWW datasets are used to test the architecture with different window sizes and prediction horizons. Experimental results indicate that modest observation windows and short-term perspectives yield the optimal correlation between response time and prediction stability. On the other hand, the hybrid ensemble always does better than single models in different types of workloads. These results show how effectively the suggested technique works for proactive provisioning of cloud resources and reliable auto-scaling in environments that are cloud-native.