| Abstract: |
Dynamic and unpredictable workload behaviour is one of the principal challenges in cloud-resource management. Conventional threshold-based autoscaling mechanisms react only after resource utilization crosses predefined limits, potentially resulting in delayed provisioning, performance degradation, over-provisioning and unnecessary energy consumption. Artificial Intelligence enables cloud systems to move from reactive to proactive resource management by predicting future workload demands and provisioning resources before demand changes occur. This paper proposes an intelligent workload-prediction and proactive autoscaling framework combining historical cloud monitoring data, machine-learning-based forecasting and adaptive resource provisioning. The framework evaluates workload trends and predicts future CPU, memory and service demand before determining virtual-machine or container scaling decisions. Long Short-Term Memory, Transformer-based forecasting and reinforcement-learning approaches are considered for workload prediction and scaling optimization. The model aims to minimize prediction error, response time, SLA violations, resource wastage, cost and energy consumption. Recent research demonstrates that prediction-based resource management is increasingly important for cloud-edge and large-scale distributed environments. The proposed framework contributes to the Ph.D. theme “AI Driven Analysis & Optimization Framework for Cloud Based Systems” by providing a predictive optimization layer capable of supporting intelligent, sustainable and cost-efficient cloud infrastructures. |
