Hybrid data attribute reduction based on neighborhood rough set

The attribute reduction in rough set theory has considerable value in areas such as pattern classification, knowledge extraction, and data mining. Hybrid information systems usually contain discrete classification attributes and continuous numerical attributes. For such data, the neighborhood rough set model is typically used to handle hybrid attributes, but using a fixed neighborhood radius may affect the classification performance of the data. Meanwhile, as an effective tool for uncertainty reasoning, fuzzy evidence theory has advantages in handling fuzzy information and has been successfully applied in scenarios such as medical diagnosis and intelligent decision-making. This study proposes a neighborhood and fuzzy rough set model that combines data features to handle hybrid data, and designs two attribute reduction algorithms based on neighborhood and fuzzy theories to obtain reduced sets that retain the original data information. Specifically, based on the features of hybrid data, fuzzy similarity relations and neighborhood similarity relations are constructed for each attribute. This algorithm effectively improves the classification performance of the data by adopting differentiated similarity radii. On this basis, a new information entropy function is defined based on neighborhood and fuzzy theories to determine the importance of attributes. Based on the above indicators, the design idea of the attribute reduction algorithm is proposed. In the algorithm implementation process, information entropy is introduced to pre-sort attributes to optimize the search efficiency of attribute subsets and prioritize features with high discrimination. Experimental comparative analysis shows that the proposed IEDC and ARFC algorithms outperform existing methods in classification performance.