基于精英个体划分的变步长萤火虫算法的特征选择方法

针对标准萤火虫算法(firefly algorithm,FA)收敛速度慢及其在解空间内的搜索易陷入局部最优的缺陷，充分考虑萤火虫算法在寻优过程中其种群内个体的差异性，提出一种基于精英萤火虫个体划分的变步长策略，改进后的FA在算法迭代中对每代目标值较好的精英萤火虫个体随机增大其移动步长，而对每代目标值较差的非精英个体则线性减小其步长。为适用于特征选择问题，又对FA中萤火虫的编码和位置移动进行了离散化定义，给出了基于所提改进型离散FA(binary firefly algorithm,BFA)的包装式特征选择方法流程。在UCI分类数据集上对比测试了所提改进型BFA与其他算法在优化特征选择方面的性能。测试结果表明，基于所提改进型BFA优化特征选择的效果较好，验证了所提改进策略可有效提升FA的优化能力。

Feature selection method based on the dynamic step firefly algorithm with the elite individual dipartition

In order to overcome the shortcomings of the firefly algorithm (FA) such as slow convergence speed and falling into local optimum of solution space, the paper fully considers the individual differences among the population of fireflies when FA is running for optimization. And a dynamic step strategy with the elite firefly individual dipartition is proposed, which randomly increases the moving step of elite firefly individuals with better target values in every algorithm iteration but linearly reduces the moving step of non-elite individuals with poor target values. The coding and positional movement of fireflies of FA are defined in binary mode for the feature selection problem, then the procedure of the wrapped feature selection method based on the improved binary FA (binary firefly algorithm, BFA) is presented. Finally,a comparative test of feature selection optimization ability between the improved BFA and other algorithms is executed on UCI classification datasets. The test result shows the improved BFA has better optimization effect on feature selection than other algorithms, verifying that the proposed improvement strategy can effectively improve the optimization ability of FA.