Neighborhood Rough Set Reduction with Improved Mayfly Optimization Algorithm

Abstract

Attribute reduction refers to selecting some of the most effective attributes from the original attribute sets to reduce the dimensionality of the dataset and optimize the attribute indicators of the system. Rough set theory has made good achievements in attribute reduction as dataset preprocessor, but the current methods are not enough to find the global best subset. The Mayfly algorithm is a new intelligent optimization algorithm, which mainly includes group gathering, crossing, mutation, wedding dance and random walking. In this paper, an attribute reduction algorithm of neighborhood rough set based on improved MA algorithm is proposed. Two Sigmoid mapping functions are used to binary encoding of the mayfly, the switching conditions of the two positions update strategies are set to control migration trend distribution of the mayfly. And then the neighborhood rough set is combined to find the best attribute subset. The UCI and OpenML datasets are used as experiment datasets. The experimental results verify the superiority of the proposed method in finding an optimal reduction. At the same time, the proposed method has higher classification accuracy and faster convergence speed.