Data centers form the backbone of the global digital economy, yet their exponential growth has led to significant energy consumption, with cooling systems alone accounting for 30 to 50% of total energy usage. This paper proposes an integrated framework that combines machine learning based workload prediction with dynamic cooling control to achieve holistic energy optimization. The system employs a multi-layered architecture comprising real-time sensor telemetry, predictive analytics (Random For-est and Reinforcement Learning agents), and adaptive actuation of both passive and active cooling technologies. A simulation environment is developed to model varying workload patterns and evaluate the impact of the proposed controller against a static baseline. Results indicate a reduction in cooling power of up to 30% while maintaining thermal safety and computational performance. The work further discusses the incorporation of sustainability metrics beyond Power Usage Effectiveness (PUE), including Water Usage Effectiveness (WUE) and Carbon Usage Effectiveness (CUE). The proposed approach demonstrates that intelligent coordination of IT and cooling resources is a viable pathway toward greener, more efficient data center operations.

Green Data Centers Cooling Optimization PUE Predictive Modeling Reinforcement Learning Resource Management