An Improved Sea Lion Optimization for Workload Elasticity Prediction with Neural Networks

Abstract

Abstract The work in this paper presents a study into nature-inspired optimization applied to workload elasticity prediction using neural networks. Currently, the trend is for proactive decision support in increasing or decreasing the available resource in cloud computing. The aim is to avoid overprovision leading to resource waste and to avoid resource under-provisioning. The combination of optimization and neural networks has potential for the performance, accuracy, and stability of the prediction solution. In this context, we initially proposed an improved variant of sea lion optimization (ISLO) to boost the efficiency of the original in solving optimization problems. The designed optimization results are validated against eight well-known metaheuristic algorithms on 20 benchmark functions of CEC’2014 and CEC’2015. After that, improved sea lion optimization (ISLO) is used to train a hybrid neural network. Finally, the trained neural model is used for resource auto-scaling based on workload prediction with 4 real and public datasets. The experiments show that our neural network model provides improved results in comparison with other models, especially in comparison with neural networks trained using the original sea lion optimization. The proposed ISLO proved efficiency and improvement in solving problems ranging from global optimization with swarm intelligence to the prediction of workload elasticity.

Keywords

• Nature-inspired computing
• Improved sea lion optimization
• Memorizing historical movement
• Levy flight trajectory
• Opposition-based learning
• Extreme learning machine