基于自学习二元差分进化的多目标特征选择

MULTIPLE OBJECTIVE FEATURE SELECTION BASED ON SELF-LEARNING BINARY DIFFERENTIAL EVOLUTION

为提升特征选择算法的搜索能力,加快收敛速度,提出一种基于自学习二元差分进化的多目标特征选择方法。引入三种算子,基于概率差的二元变异算子来产生最优解,从而快速地引导个体定位潜在的最优区域。另外,引入的净化搜索算子可以提高处于最优区域的精英个体的自学习能力,而具有拥挤距离的非支配排序算子可以降低差分进化中选择算子的计算复杂度。在多个数据集的实验结果表明,提出的方法能够实现高效精确的多目标特征选择。

In order to improve the group search ability and accelerate the convergence speed,a multiple objective feature selection method based on selflearning binary differential evolution is proposed.Three operators were introduced,and the binary mutation operator based on probability difference was used to generate the optimal solution,so as to quickly guide individuals to locate the potential optimal region.The clean search operator was introduced to improve the selflearning ability of the elites in the optimal region,while the nondominated sorting operator with crowding distance could reduce the computational complexity of the selection operator in differential evolution.The experimental results on multiple data sets show that the proposed method is efficient and accurate on multiple objective feature selection.