Attribute reduction in an incomplete categorical decision information system based on fuzzy rough sets

Categorical data is an important class of data in machine learning. Information system based on categorical data is called a categorical information system (CIS), a CIS with missing values is known as an incomplete categorical information system (ICIS) and an ICIS with decision attributes is said to be an incomplete categorical decision information system (ICDIS). Attribute selection is an important subject in rough set theory. This paper investigates attribute reduction in an ICDIS based on fuzzy rough sets. To depict the similarity for incomplete categorical data, fuzzy symmetry relations in an ICDIS are first introduced. Then, some attribute-evaluation functions, such fuzzy positive regions, dependency function and attribute importance functions are given. Next, the fuzzy-rough iterative computation model for an ICDIS is presented, and an attribute reduction algorithm in an ICDIS based on fuzzy rough sets is given. Finally, experiments are carried out as so to evaluate the performance of the proposed algorithm, and Friedman test and Bonferroni-Dunn test in statistics are conducted. The experimental results indicate that the proposed algorithm is more effective than some existing algorithms.

Automated summary

This paper investigates attribute reduction in an incomplete categorical decision information system (ICDIS) based on fuzzy rough sets. An attribute reduction algorithm is proposed and experiments show that it outperforms existing algorithms.