Dynamic graph-based attribute reduction approach with fuzzy rough sets

Lei Ma, Chuan Luo, Tianrui Li, Hongmei Chen, Dun Liu

Abstract: Incremental datasets are becoming increasingly common as interesting data are continually accumulated across various application fields. Selecting informative attributes from dynamically changing datasets poses numerous challenges. Completely reapplying the attribute reduction algorithm to detect the changes in the data and learn the selected attributes following frequently changing data is prohibitively expensive. In this regard, an incremental processing mechanism is desired to facilitate progressively updating the attribute reducts when the data is updated. In this paper, we consider the maintenance of the fuzzy rough attribute reduction in dynamic data that is changing through the arrival of samples. Based on the transformation of attribute reduction in a fuzzy decision system into the minimal transversal of a derivative hypergraph, a novel dynamic fuzzy rough attribute reduction approach is presented from a graph-theoretic perspective, so as to facilitate efficient computation of reduct in incremental datasets. Extensive experimental evaluation shows that the proposed dynamic graph-based fuzzy rough approach provides significantly faster attribute reduction than completely re-reduction by its original static counterpart as well as the existing dynamic attribute reduction approach based on fuzzy discernibility matrix, and is also effective in preserving the quality of the selected reduct.