Attribute group for attribute reduction

Abstract In the field of rough set, how to improve the efficiency of obtaining reduct has been paid much attention to. One of the typical strategies is to reduce the number of comparisons among samples, it follows that the time consumption of generating binary relation can be saved for quickly obtaining reduct. Nevertheless, such mechanism is only designed by reducing the number of times to scan samples, which fails in reducing the iterations of evaluating attributes. To fill such gap, an acceleration strategy based on attribute group is proposed. Firstly, all of the candidate attributes are divided into different groups. Secondly, in the process of searching reduct, only the attributes out of those groups which contain at least one attribute in the potential reduct should be evaluated. It should be noticed that this is the key which can reduce the number of evaluations of candidate attributes. Finally, the above two steps are repeated until the constraint defined in attribute reduction is satisfied. To demonstrate the effectiveness of our proposed method, the experiments have been conducted over three neighborhood rough set models by leveraging four measures. Compared with the existing forward greedy searching approach over 12 UCI data sets, the experimental results show that our attribute group based approach can maintain the classification performance derived by reducts while significantly accelerating the searching process of obtaining reduct. This study suggests a new trend concerning the problem of quickly computing reduct.