Published Paper Details:

Cloud computing environments increasingly support cloud-native and microservices-based applications that exhibit highly dynamic and burst-prone workload patterns. Conventional threshold-based autoscaling mechanisms are inherently reactive and frequently suffer from delayed resource provisioning, poor utilization efficiency, and Service Level Agreement (SLA) violations during sudden demand fluctuations. To overcome these limitations, this paper proposes an AI-driven predictive analytics framework for proactive cloud resource management that integrates time-series workload forecasting with cost-aware optimization. The proposed framework employs Long Short-Term Memory (LSTM) networks to predict short-term workload variations and a reinforcement learning (RL)-based decision engine to derive adaptive autoscaling policies that jointly optimize performance, resource utilization, and operational cost under SLA constraints. Unlike existing approaches that address workload prediction or cost optimization in isolation, the proposed model enables closed-loop, multi-objective optimization through continuous feedback-driven learning. The framework is evaluated using benchmark cloud workload traces and a production-like containerized cloud deployment. Experimental evaluation is conducted using publicly available benchmark workload traces and a containerized cloud testbed emulating real-world deployment conditions. Results averaged over multiple execution runs demonstrate that the proposed approach reduces average response time by 25-30%, demonstrates resource utilization by 18-22%, and lowers operational cost by 15-18% compared to conventional threshold-based autoscaling, while maintaining SLA compliance above 98%. These results confirm that integrating predictive intelligence with adaptive decision-making significantly enhances scalability, efficiency, and cost-effectiveness in modern cloud environments.