Optimized Resource Allocation for High-Performance Cloud Systems using Machine Learning

Cloud computing forms the foundation of modern digital infrastructure, especially in multi-cloud environments where optimized resource allocation is critical for ensuring performance, reliability, and low-latency service delivery. This research aims to fill the gap in dynamic, intelligent resource allocation strategies by addressing issues like high-dimensional feature noise, data drift, and workload variability—limitations often overlooked in traditional rule-based approaches. A novel machine learning pipeline is proposed, integrating Leaf-Wise Feature Bundling (LFB) and Adversarial Validation Re-weighting into a LightGBM classifier. LFB reduces feature dimensionality by clustering highly correlated attributes, while adversarial validation detects train-test distribution shifts and assigns sample-specific weights to enhance generalization. The combined approach significantly improves classification performance, achieving 100% accuracy on a real-world multi-cloud dataset and outperforming baseline models. Experimental analyses confirm the robustness, scalability, and adaptability of the model, highlighting its potential for intelligent, real-time decision-making in complex, high-performance cloud systems.