Improvement on classification performance based on multiple reduct ensemble
Qinghua Hu, Dan Yu, Zongxia Xie · 2004
Rough set approaches are widely applied to feature selection and data mining. The minimal reduct of an information system is preferred in traditional rough set approaches according to minimal description length principle. In this paper, we present some experiments and find a minimal reduct is a weaker solution for the given classification task in most cases. A multiple classifier system based on rough set reduction is proposed, which improves the performance by combining multiple rough set based reducts. Experiments based on CART and SVM show the proposed method is efficient and effective