Revolutionizing Autonomous Cloud Infrastructure: AI-Driven Predictive Auto Scaling with Attribute-Based Instance Selection in AWS

Dynamic resource provisioning is essential for cost efficiency and performance in cloud computing, yet prevailing auto-scaling practices are predominantly reactive. This paper presents a novel framework that integrates advanced predictive analytics—employing a hybrid of LSTM and Transformer-based models—with Amazon EC2’s attribute-based instance selection in Auto Scaling Groups. Our system learns from 90 days of multi-resolution workload data and leverages adaptive statistical confidence metrics to adjust pricing thresholds for Spot Instances. Simulated experiments using real-world AWS workload traces demonstrate that our approach reduces scaling latency by 75%, improves resource utilization by 20–30%, and lowers costs by 35% compared to conventional threshold-based methods (p < 0.001). Additionally, a rigorous sensitivity analysis of key scaling parameters confirms the robustness of the proposed framework.