A Robust Fuzzy Rough Set Method Based on Multi-Kernel and Fuzzy Decision for Feature Selection

Abstract

Rough set is a commonly used feature selection tool. Fuzzy rough sets further expand the adaptability of the model. Unfortunately, noise can have a large effect on the calculation results of fuzzy rough sets. Such sensitivity severely limits the practical applications of rough fuzzy rough sets. In order to solve the above problems, some powerful robust models had been proposed in recent years, but most of these methods identify the noisy samples in a certain way and then ignore them, which will lead to the loss of information. Therefore, in this work, a robust fuzzy rough set model based on multi-kernel and fuzzy decision was proposed. This method introduces the concept of fuzzy decision to compute the membership degree of each sample for each decision attribute, which can initially eliminates the impact of manual misclassification. Secondly, the multi-kernel operators were used to measure the similarity between samples, so as to improve the ability of the model to deal with nonlinear classification problems. Finally, the k-nearest neighbor idea was used to further weaken the error caused by the noisy samples. Further, a feature selection strategy based on greedy algorithm is proposed. Ten data sets were selected from UCI to compare the proposed algorithm with some of the state-of-the art approaches. The comparison results of this experiment prove that the proposed feature selection algorithm is more effective in most cases when selecting some specific multi-kernel operators.