A graph approach for fuzzy-rough feature selection

Rough sets, especially fuzzy-rough sets, have proven to be a powerful tool for dealing with vagueness and uncertainty in data analysis. Fuzzy-rough feature selection has been shown to be highly useful in data dimensionality reduction. However, many fuzzy-rough feature selection algorithms are still time-consuming when dealing with the large-scale data sets. In this paper, the problem of feature selection in fuzzy-rough sets is studied in the framework of graph theory. We propose a new mechanism for fuzzy-rough feature selection. It is shown that finding the attribute reduction of a fuzzy decision system can be translated into finding the transversal of a derivative hypergraph. Based on the graph-representation model, a novel graph-theoretic algorithm for fuzzy-rough feature selection is proposed. The performance of the proposed method is compared with those of the state-of-the-art methods on various classification tasks. Experimental results show that the proposed technique outperforms all other known feature selection methods in terms of the computation time. Especially for the large-scale data sets, it demonstrates promising performance. Moreover, our proposed method can achieve better classification accuracies with the usage of small number of features.