An Optimized Deep Learning Model for Predicting Workload in Large Scale Cloud Platforms

Cloud workloads are inherently dynamic, influenced by unpredictable user behavior, seasonal traffic variations, and sudden spikes in demand caused by events such as sales campaigns or system updates. Traditional statistical prediction models often struggle to capture these rapidly changing patterns. Without accurate workload forecasting, cloud service providers risk under- provisioning—leading to service delays and downtime—or over-provisioning, which increases operational costs. This unpredictability makes intelligent forecasting a necessity. Accurate workload prediction enables cloud platforms to allocate resources dynamically and proactively. This improves server utilization, reduces idle time, and ensures that computational power is available when needed, without excessive redundancy. For enterprises, this means lower operational costs, while for service providers, it results in better infrastructure. However, due to the absence of seasonality in the data patterns coupled with the sporadic nature of cloud workload, accurate prediction is a challenge. This paper presents a deep learning based approach with data optimization for predicting cloud workloads. It has been shown that the proposed approach clearly outperforms existing approaches in terms of prediction accuracy. Keywords—Cloud Workload Estimation, Deep Neural Network (DNN), Steepest Descent Approach, Mean Absolute Percentage Error (MAPE).