Incremental attribute reduction for dynamic fuzzy decision information systems based on fuzzy knowledge granularity

Rough set-based attribute reduction is a powerful technique for data preprocessing in data mining. Knowledge granularity, as a reliable measure for assessing uncertainty in decision information systems (DIS), finds applicability in attribute reduction within such systems. Nevertheless, the limitation arises from the fact that static attribute reduction methods fail to effectively utilize the information contained in acquired data and promptly update knowledge due to the continuous evolution of data. In addition, existing incremental methods based on knowledge granularity are designed exclusively for symbolic data and lack the capability to handle real-valued data. Inspired by this, our study focuses on the attribute reduction approach for fuzzy decision information systems (FDIS) that encompass object variations by utilizing fuzzy knowledge granularity. Firstly, fuzzy knowledge granulation is defined to quantify uncertainty within FDIS, and utilized to determine the importance of attributes for attribute reduction. Additionally, the incremental mechanisms and attribute reduction algorithms are investigated for adding an object and an object set to FDIS, respectively. Moreover, an explication of how the incremental mechanism for increasing an object set can be viewed as a generalization of the mechanism used for a single object is provided. Finally, comparative experiments on various datasets are conducted to validate the effectiveness and efficiency of the proposed incremental algorithms. The results demonstrate that our algorithms achieve superior classification accuracy and while requiring minimal computing time when compared to the comparative algorithms.