Load balancing for cloud computing using optimized cluster based federated learning

Task scheduling and load balancing in cloud computing represent challenging NP-hard optimization problems that often result in inefficient resource utilization, elevated energy consumption, and prolonged execution times. This study introduces a novel Cluster-based Federated Learning (FL) framework that addresses system heterogeneity by clustering virtual machines (VMs) with similar characteristics via unsupervised learning, enabling dynamic and efficient task allocation. The proposed method leverages VM capabilities and a derivative-based objective function to optimize scheduling. We benchmark the approach against established metaheuristic algorithms including Whale Optimization Algorithm (WOA), Butterfly Optimization (BFO), Mayfly Optimization (MFO), and Fire Hawk Optimization (FHO). Evaluated using makespan, idle time, and degree of imbalance, the Cluster-based FL model coupled with the COA algorithm consistently outperforms existing methods, achieving up to a 10% reduction in makespan, a 15% decrease in idle time, and a significant improvement in load balancing across VMs. These results highlight the efficacy of integrating clustering within federated learning paradigms to deliver scalable, adaptive, and resilient cloud resource management solutions.