A Weighted Granular-Ball Rough Set: Model and Attribute Reduction

The granular-ball rough set (GBRS) is an effective rough set in recent years that unifies the classical rough set and the neighborhood rough set. However, it does not take into account the correlation between conditional attributes and decision attribute. To fill such a gap, we consider the weight of conditional attributes, the weight of a conditional attribute reflects its contribution to the decision attribute. By assigning appropriate weights, the relationship between each conditional attribute and decision attribute can be accurately evaluated, thereby improving the predictive accuracy and applicability of the model. To further reduce the dimensionality of the data, a greedy searching algorithm based on weighted granular-ball rough set (WGBRS) is designed to select a subset of conditional attributes with both strong correlation and high dependency. Finally, ten datasets from the UCI machine learning repository are used to compare the performance of the attribute reduction algorithm based on WGBRS with the other three popular attribute reduction algorithms. Experimental results show that WGBRS has high classification accuracy and favorable attribute reduction effects.