Quickly calculating reduct: An attribute relationship based approach

Presently, attribute reduction, as one of the most important topics in the field of rough set, has been widely explored from different perspectives. To derive the qualified reduct defined in attribute reduction, forward greedy searching is frequently used. However, the previous researches indicate that such searching strategy may be still computationally expensive if the volume of data is large. In view of this, two frameworks are proposed by considering the relationships between attributes, which aim to accelerate the process of searching reducts. Our consideration is actually realized based on the dissimilarity and similarity between attributes, respectively. The main mechanisms are: (1) for the dissimilarity based approach, the combination of attributes with significant difference instead of one and only one attribute will be added into potential reduct in the process of searching reduct; (2) for the similarity based approach, the candidate attributes which are similar to those attributes in potential reduct will be tentatively ignored instead of being evaluated in the process of searching reduct. The experimental results over 16 UCI data sets demonstrate that whether single granularity or multi-granularity attribute reduction is considered, our proposed approaches can not only generate the reducts which may not lead to poorer performances, but also provide superior time efficiency of calculating reducts. This study suggests new trends for quickly computing reducts. •An acceleration mechanism for attribute reduction is considered from the perspective of attribute relationship.•The dissimilarity and similarity between attributes are introduced into the framework of quickly calculating reducts.•Our proposed approaches can reduce the elapsed time of deriving reduct.