A new scalable and performance-enhancing bootstrap aggregating scheme for variables selection: Taking real-world web services resources as a case

Abstract

Variables selection is a vital Data Mining technique which is used to select the cost-effective predictors by discarding variables with little or no predictive power. In this paper, we introduce a new conceptual model for variables selection which includes subset generation, Ensemble learning, models selection and validation. Particularly, we addressed the problem of searching for and discarding irrelevant variables, scoring variables by relevance and selecting a subset of the cost-effective predictors. The generalization was seen to improve significantly in terms of recognition accuracy when the proposed system, which is named SPAS, is tested on QoS for Real-World Web Services. Good experimental studies demonstrate the effectiveness of our Wrapper model.