A fuzzy rough set approach for incremental feature selection on hybrid information systems

In real-applications, there may exist many kinds of data (e.g., boolean, categorical, real-valued and set-valued data) and missing data in an information system which is called as a Hybrid Information System (HIS). A new Hybrid Distance (HD) in HIS is developed based on the value difference metric, and a novel fuzzy rough set is constructed by combining the HD distance and the Gaussian kernel. Considering the information systems often vary with time, the updating mechanisms for attribute reduction (feature selection) are analyzed with the variation of the attribute set. Fuzzy rough set approaches for incremental feature selection on HIS are presented. Then two corresponding incremental algorithms are proposed, respectively. Finally, extensive experiments on eight datasets from UCI and an artificial dataset show that the incremental approaches significantly outperform non-incremental approaches with feature selection in the computational time.