A Feature Selection Method for Online Hybrid Data Based on Fuzzy-rough Techniques

Data reduct based on rough set theory was an effective feature selection method, however, classic rough set theory cannot deal with hybrid data and canpsilat applied to online systems either. So the rough set model based on fuzzy equation relation was improved to reduct the hybrid systems. The entropy was used to measure the discernibility power of the information and the definition of relative reduct was improved, and the notion of sequential reduct was proposed to deal with real online systems. A complete algorithm was proposed and applied to several UCI data. Experiments show that sequential reduct algorithm is an effective feature selection method for real online systems.