Dynamic resource management has emerged as a critical requirement in hyperscale cloud environments, where escalating workload volatility, architectural heterogeneity and multi-tenant interference continuously undermine the reliability of conventional provisioning mechanisms. This systematic review consolidates and evaluates state-of-the-art methodologies by following PRISMA-2020 guidelines and extracting evidence exclusively from Q1 and SCI-indexed literature, thereby ensuring methodological robustness and empirical validity. A structured search strategy and a multi-dimensional quality-assessment matrix were employed to filter, classify and synthesize studies based on experimental rigor, reproducibility, dataset transparency and statistical soundness. The resulting taxonomy has incorporated heuristic and meta-heuristic optimizers, machine-learning-based predictors, reinforcement-learning controllers, control-theoretic models, game-theoretic models, and constrained optimization solvers, thus allowing a common understanding of their operation behaviour, scalability properties and applicability to VM-based and container-native infrastructures. Comparative synthesis reveals that while heuristic and control-theoretic methods excel in low-latency responsiveness, machine-learning and reinforcement-learning models provide superior predictive and adaptive capabilities, yet suffer from training overheads, concept drift and stability issues in highly dynamic settings. Persistent challenges including forecasting uncertainty, inconsistent benchmarking, energy SLA trade-offs, migration overheads and security vulnerabilities continue to limit the generalizability of existing solutions. The review also establishes recent trends, like serverless and function-level autoscaling, edge- cloud continuum integration, AI-native autonomous control planes, carbon-aware workload placement, and quantum-inspired optimization, which are expected to define the future development of next-generation resource- orchestration systems. Overall, this review underscores the necessity for hybrid, multi-layered decision architectures, standardized evaluation pipelines and trustworthy, self-optimizing control loops capable of sustaining elasticity, efficiency and resilience in future cloud ecosystems.