Intelligent Load Balancing in Cloud Computing Using Machine Learning Algorithm’s for Adaptive Resource Allocation
Devendra Namdeo, Dr. Jyotibala Gupta, Dr. Praveen Kumar Shrivastava, Dr. Ayush kumar Agrawal, Dr. Abhinav Shukla
Cloud computing platforms must continuously accommodate workloads that vary in intensity, duration, resource requirements, and quality-of-service (QoS) expectations. Conventional load-balancing and resource-allocation mechanisms commonly depend on static thresholds, instantaneous utilization, or manually configured policies. Such approaches can react slowly to workload changes and may produce either over-provisioning or resource contention. This paper proposes an Intelligent Machine Learning-based Load Balancing and Adaptive Resource Allocation (IML-ABARA) framework that combines workload prediction, resource-state estimation, adaptive load scoring, and policy-based resource allocation. The framework is designed to predict near-future workload demand and use the prediction together with current CPU, memory, network, queue, latency, and task-level information to select suitable compute resources. A hybrid architecture is proposed in which supervised learning performs short-horizon workload forecasting while a reinforcement-learning-inspired decision layer selects allocation actions under multi-objective constraints. The optimization objective jointly considers response time, resource utilization, load imbalance, energy-related cost, SLA violations, and scaling cost. A reproducible experimental protocol is defined using CloudSim/CloudSim Plus or an equivalent containerized testbed, with baseline comparison against Round Robin, Weighted Round Robin, Least Connections/Least Loaded, threshold-based autoscaling, and a conventional machine-learning load balancer. The paper contributes a formal problem formulation, feature engineering strategy, adaptive load-score model, decision policy, evaluation protocol, and deployment architecture. Importantly, numerical performance claims are not fabricated; the manuscript specifies the experiments required to validate the proposed framework. The approach is intended to provide a practical foundation for predictive and adaptive cloud resource management in heterogeneous and elastic environments.