The foremost challenge in cloud computing is managing the quality of service (QoS) at reasonable price. This work present a new auto-scaling methodology that dynamically scales cloud resources in real time based on the requirements of Virtual Network Functions (VNFs). Real-time processing in the suggested system avoids inefficiency of conventional schedule-based and threshold-based scaling mechanisms, where in most scenarios the performance will be inferior due to outdated rules or incorrect configurations. To address these limitations, our approach forecasts future CPU, memory, and bandwidth requirements of Service Function Chains (SFCs) based on a hybrid deep learning model of Multilayer Perceptron (MLP) and Long Short-Term Memory (LSTM) networks. Wireless Sensor Networks (WSNs) also enhance predictive accuracy with real-time system data. The system employs an auto-regression model to predict variable resource requirements in real-time operations and utilizes auto correlation for anomaly detection based on pattern recognition in uncertain cloud environments. This method ensures effective resource usage, system scalability, and optimal service provision based on improved demand forecasting. The proposed method is an ideal candidate for managing dynamic workloads in cloud computing as it properly balances cost reduction with performance enhancement.