Energy-Efficient Resource Allocation in Cloud Service Management via Significant Feature Ranking With Recurrent Network Model

Abstract

The increasing adoption of cloud computing (CC) produces a notable growth in data centres, resulting in severe financial, environmental, and operational problems. To mitigate these issues, energy-efficient resource allocation strategies have been developed as a key solution for enhancing energy utilization while ensuring high performance and reliability. In this framework, machine learning (ML) enables the implementation of dynamic and intelligent approaches for effective resource utilization. In this paper, we focus on the Scalable Hybrid Feature Ranking Framework for Energy-Efficient Resource Allocation (SHFRF-EERA) approach in Cloud Service Management Systems. The novelty of the proposed model lies in the integration of hybrid feature selection with an attention-based VAE-LSTM architecture optimized using AMSGrad for efficient cloud resource allocation. This combination enables improved prediction accuracy and better handling of complex workload patterns compared to existing models. Initially, data pre-processing is performed by handling missing values and data scaling techniques to ensure data consistency and improve model performance. Subsequently, a hybrid recursive backward elimination method is employed for effective feature selection, reducing dimensionality but maintaining the most appropriate attributes. For resource allocation classification process, an attention-based autoencoder integrated with recurrent network can be exploited to capture both latent representations and temporal dependencies in cloud workload patterns. Finally, AMSGrad optimization algorithm has been applied to enhance convergence stability and improve learning efficiency. To validate the efficacy of the proposed SHFRF-EERA technique, a series of simulations is conducted on the benchmark Energy-Efficient Cloud Resource Allocation dataset. The comparative result analysis demonstrates the promising performance of the proposed SHFRF-EERA method achieving an accur_y of 96.15%, preci_n of 90.62%, recal_l of 86.50%, and F1score of 88.15%. Therefore, the proposed model is found to a robust, scalable, and efficient tool for cloud resource management.