Publication

Abstract : Efficient cloud resource allocation through timeseries forecasting is critical in cloud data centers, as accurate demand prediction enables optimal utilization of computing resources and cost reduction. Traditional methods based on conventional machine learning or statistical analysis often fall short in capturing complex temporal patterns, resulting in low prediction accuracy and inefficient resource use. To address these challenges, this study introduces a novel Cycle-Consistent Adversarial Adaptation Network with Geyser Inspired Algorithm (CCAAN-GIA), leveraging historical workload data from Bitbrain's cloud data center, specifically from virtual machines in the fast Storage and Rnd datasets. The raw data undergoes pre-processing; including handling of missing values, feature averaging, and min-max normalization to prepare it for time-series forecasting using the CCAAN framework. The model's predictive performance is further enhanced by tuning hyper parameters with the Geyser Inspired Algorithm (GIA). Additionally, to optimize resource distribution and minimize operational costs, the Cat Optimization Algorithm (COA) is employed, ensuring effective resource management. Experimental results demonstrate the superiority of the proposed approach, achieving 99.5% accuracy with significantly lower error metrics-MAE of 0.02, RMSE of 11.2, and MAPE of 0.46-outperforming traditional forecasting models. This confirms the method's potential as a robust solution for intelligent and cost-efficient cloud resource allocation.