OGUARE GIFT ONOFU

Cloud computing has become key to modern digital infrastructure, yet traditional reactive auto-scaling systems struggle to balance performance requirements with cost efficiency. This study addresses the limitations of existing cloud resource management approaches by developing the Predictive and Cost-Optimized Auto-Scaling Framework (PCOAF), a conceptual model that integrates machine learning-based workload forecasting with multi- objective optimization. Through systematic application of Design Science Research Methodology, the research analyzed current auto-scaling systems, identified critical deficiencies including reactive latency, prediction inaccuracy, and cost inefficiency, and designed a three-layered architecture comprising monitoring, prediction and decision, and optimization and execution modules. The framework employs archetype-aware prediction to classify workloads into four behavioral patterns; SPIKE, PERIODIC, RAMP, and STATIONARY enabling tailored scaling strategies for each type. Theoretical validation demonstrates that PCOAF achieves 99.8% workload classification accuracy, reduces mean absolute percentage error to 15%, and projects cost reductions of 22% while decreasing service-level objective violations by 61.4% compared to baseline reactive systems. The study establishes PCOAF's feasibility across five design criteria: relevance, consistency, feasibility, scalability, and economic viability. By addressing identified gaps in both international research and Nigeria's emerging cloud ecosystem, this framework contributes a theoretically grounded and practically applicable solution for intelligent, cost-aware cloud resource management in resource-constrained environments.