An ant colony optimization attribute reduction algorithm for hybrid data using fuzzy β covering and fuzzy mutual information

Abstract

As an effective tool for handling the uncertainty and fuzziness of data, fuzzy $\beta$ covering can fit the given dataset well. Swarm intelligence algorithms are suitable for solving complex combinatorial optimization problems and then have unique advantages in attribute reduction. This paper proposes an ant colony optimization attribute reduction algorithm based on fuzzy $\beta$ covering and fuzzy mutual information. Initially, a fuzzy $\beta$ covering decision information system for hybrid data is built based on fuzzy $\beta$ covering theory. Then, fuzzy mutual information is introduced to measure the uncertainty of this system. Subsequently, an evaluation function is constructed using fuzzy mutual information for designing a forward attribute reduction algorithm based on heuristic search strategy. Moreover, to identify potentially more optimal attribute subsets, an ant colony optimization attribute reduction algorithm based on random search strategy is designed. Finally, two proposed algorithms are experimentally compared with six existing attribute reduction algorithms. The results indicate that these two algorithms surpass the other six algorithms in terms of classification accuracy and reduction rate.