MACHINE LEARNING-BASED NETWORK INTELLIGENCE FRAMEWORK FOR AUTONOMOUS DATA CENTER OPERATIONS

Data centers increasingly depend on highly dynamic, software-defined networks whose operating conditions change faster than manual procedures can safely accommodate. This review examines how machine learning can be organized as a network intelligence layer for autonomous data center operations, with emphasis on telemetry interpretation, anomaly detection, forecasting, graph-based network modeling, reinforcement learning, intent translation, and closed-loop assurance. An integrative review approach is used to synthesize peer-reviewed research published primarily from 2020 to 2025 across networking, cloud computing, data - center energy management, artificial intelligence, and autonomous systems. The evidence indicates that no single algorithm is sufficient for end-to-end autonomy. Supervised and unsupervised models are effective for classification and anomaly recognition, graph neural networks improve representation of topology-dependent behavior, forecasting models anticipate load and congestion, and reinforcement learning supports sequential control for routing, consolidation, and cooling. However, operational deployment remains constrained by concept drift, incomplete observability, sparse labels, unsafe exploration, weak cross - domain coordination, and limited evidence from production - scale evaluations.