Hybrid Deep Learning Framework for Cloud Workload Prediction: Bridging Accuracy and Scalability

Abstract

Cloud computing provides the services in the online according to the availability of the services. Cloud workloads are anticipated in a rapidly evolving cloud computing environment to perform optimal resource allocation, increase performance, and decrease costs. Traditional deep learning (DL) systems have difficulty balancing accuracy vs. scalability trade-offs in dynamic cloud environments that host diverse and unpredictable workloads. We introduce a Hybrid Deep Load Learning Framework (HDLLF) that merges Weighted Least Connections (WLC) with Adaptive Backpropagation Neural Network (BPNN) to tackle these issues. The proposed architecture combines the advantages of WLC to extract spatial features and BPNN to extract temporal patterns for cloud workload prediction. By integrating both methodologies, the framework enhances prediction accuracy and maintains scalability for large-scale cloud workloads. We showcase the efficacy of our framework with extensive experiments on real-world cloud datasets, illustrating its ability to cope with diverse workloads and consistently strong performance.