A feature selection method combining neighborhood rough set and butterfly optimization algorithm

Traditional rough set based feature selection methods are difficult to obtain optimal feature subsets and require a lot of computing time. The rough set model cannot well handle continuous type and high-dimensional complex data. To tackle these problems, a feature selection method based on neighborhood rough set and improved butterfly optimization algorithm is proposed. First, the improved butterfly optimization algorithm is presented. The exploration capability of the algorithm is enhanced by introducing a sine cosine function and a nonlinear control factor to avoid the butterfly optimization algorithm from falling into local optima. A dynamic switching probability strategy is introduced to adjust the weighting of the global and local search of the butterfly optimization algorithm to reflect a reasonable search process. Then, a fitness function is constructed to evaluate and rank the fitness of the initialized feature subsets by combining the neighborhood attribute dependency and the number of features. Finally, an improved butterfly optimization algorithm is used to find an acceptable subset of near-optimal features through continuous iterations. Experiments are conducted on UCI datasets and the results show that the proposed algorithm can effectively select feature subsets with a smaller number of features and higher classification accuracy.