AI-Driven Resource Management in Distributed Systems: Predictive Accuracy and Comparative Evaluation

The rapid expansion of large-scale distributed computing environments, including cloud and hybrid edge infrastructures, has intensified the need for intelligent, adaptive, and energy-efficient resource management strategies. This paper presents an AI-driven framework that integrates Deep Neural Networks (DNN) for predictive workload forecasting, Reinforcement Learning (RL) for adaptive scheduling, and an evolutionary optimization layer for global resource equilibrium. The proposed system was evaluated across multiple scenarios to assess predictive accuracy, learning behavior, adaptability under dynamic conditions, and scalability. Experimental results demonstrate that the DNN module achieves high predictive precision for CPU and memory utilization (RMSE 0.96), while the RL agent converges to a stable, high-performing policy, improving scheduling efficiency by over 22% compared to traditional heuristics. Under dynamic disturbances such as workload surges, node failures, and network fluctuations, the framework exhibits rapid adaptation, high resource reallocation efficiency (>90%), and minimal energy overshoot (