AI Powered Hybrid Framework for Efficient VM Auto-Scaling in Cloud Computing

Virtual machine (VM) auto-scaling helps manage cloud resources while ensuring performance and cost efficiency. The rule-based and heuristic approaches fail to adapt quickly to workload changes which leads to suboptimal resource management. This research proposes an auto-scaling framework merging CNN-LSTM forecasting with Kalman filter real-time adjustment capabilities. The framework uses core workload indicators to forecast future VM requirements from analyzing memory utilization, CPU usage, traffic, disk I/O and the request rate. The Kalman filter enhances prediction accuracy by continuously adapting to sudden changes in demand levels. The framework is validated using CloudSim Plus-6 to evaluate its performance against BiLSTM, TCN, N-BEATS-GRU and Transformer models using a synthetic dataset which was created through CloudSim Plus 6. The CNN-LSTM model achieved the highest accuracy (96.66 %) with the lowest inference latency. The proposed framework integrates forecasting and correction steps into a single real-time VM scaling process which differs from current methods that handle these steps independently. This integration results in superior resource utilization, better workload response capabilities and reduced costs offering a robust auto-scaling solution for modern cloud systems.