A Deep Learning-Based Workload Forecasting Model in Cloud Data Centers
A. Prasanth, Kosuri Naresh Babu, P. Rahul, Bheemidi Vikram Reddy · 2025
Strong workload prediction in cloud data centers is necessary for efficient and reliable operations. Forecasting models, particularly the combination of Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) networks, however, are still susceptible to adversarial attacks. Strategically small perturbations can cause major distortions in prediction accuracy, making resource allocation inefficient and possibly resulting in service interruption. This research investigates such vulnerabilities with white-box adversarial attacks against CNN-LSTM models trained on actual workload data streamed from IoT sensors. A mixed methodology with MATLAB and Python processes data preprocessing, model construction, and attack simulation. Experimental outcomes demonstrate significant performance degradation under attack, which underscores the necessity of more robust defense mechanisms. Improving these models against adversarial attacks can increase cloud infrastructure robustness which guarantees stable and efficient performance in dynamic conditions.