Attribute Reduction for Heterogeneous Data by Hybrid Neighborhood Graph Structure and Neighbor Inconsistent Pair Selection

As an effective means of data processing, the rough set model simulates human cognition by partitioning objects into multiple granules. Currently, the rough set model is widely used in attribute reduction. However, further research is needed to develop rough set attribute reduction methods capable of handling heterogeneous data containing both symbolic and numerical attributes. Neighborhood rough set model helps with heterogeneous data, yet its heuristic attribute reduction methods struggle with slow computation. Furthermore, the non-transitivity of the neighborhood rough set model prevents direct inference of consistency relations between object pairs. To address these challenges, we propose a novel heterogeneous neighborhood rough set model. This model tackles the transitivity problem by introducing an innovative concept named hybrid neighborhood graph structure (HNG). The HNG represents a new granular structure, through which we redefine the concepts of neighbor consistent pairs and neighbor inconsistent pairs within the model. Additionally, the traditional heuristic attribute reduction process requires calculating the corresponding relation of all remaining attributes with each selection. In contrast, our neighbor inconsistent pair selection strategy only needs to compute the neighbor inconsistent pairs once, significantly reducing the time consumption of the attribute reduction algorithm. Furthermore, we develop two efficient attribute reduction algorithms based on neighbor inconsistent pairs to enhance computational speed and accuracy. Compared to several classical algorithms, our method demonstrates superior performance in classification accuracy and computational efficiency.