Enhanced Hybrid Model for Efficient Task Scheduling in Cloud Computing Using Deep Reinforcement Learning and Feed Forward Neural Network

ABSTRACT The dynamic, uncertain, and heterogeneous nature of cloud computing presents significant challenges in task scheduling, especially for multimedia applications with high processing requirements. Proper scheduling is essential to optimize energy consumption, makespan, and minimize Service Level Agreement (SLA) violations. Existing heuristic approaches often fail to handle heterogeneous tasks with varying CPU and memory requirements while meeting real‐time QoS constraints. To address these issues, this work proposes an Enhanced Hybrid Model integrating Q‐learning with a Feed Forward Neural Network (FNN), further optimized using the Addax optimization algorithm for hyperparameter tuning. The model evaluates Q‐values to guide task scheduling decisions based on system states, actions, and rewards. Simulation results demonstrate the effectiveness of the proposed approach, achieving a makespan of 50 s, energy consumption of 0.2 J, SLA violation of 2%, throughput of 27 tasks/s, response time of 7 s, and task prioritization of 80,000, outperforming existing methods. These results indicate that the model is adaptive, scalable, and suitable for multimedia‐based dynamic cloud environments.