Cloud Infrastructure Resource Demand Prediction Model Using Parameter Optimization and Feature Selection

Abstract

Cloud computing offer highly scalable, and economical infrastructure for promising heterogeneous platforms and various applications. According to the growing demand nature of cloud infrastructure resources, the cloud providers face the challenge of performing the effective resource management. This paper presents the development of the CPU resource demand prediction model for cloud infrastructure to overcome the critical issue of the cloud providers for workload forecasting and optimal resource management. The model is developed based on the powerful machine learning technique, Random Forests (RF) algorithm via the real data center workload traces. To get the best prediction model by RF, the parameter optimization is performed. Moreover, some features of workload traces cannot influence in prediction and also give overheads in model development time. So, the feature selection is applied to extract the important features. The performance evaluation of the proposed model against four workload traces is also presented.