基于粗糙集与人工蜂群算法的动态特征选择

Feature selection of dynamic data stream based on rough set and artificial bee colony algorithm

为提高动态数据流特征提取的计算效率与性能,设计一种基于粗糙集与人工蜂群算法的动态数据流特征选择算法。修改人工蜂群算法中雇佣蜂阶段与侦查蜂阶段的位置更新方程,降低人工蜂群算法早熟收敛的几率,增强人工蜂群算法的鲁棒性,使其满足动态特征选择算法的稳定性需要。使用粗糙集定义数据流增量数据的适应度函数,人工蜂群算法从旧特征子集与增量数据提取新的全局特征子集。基于10个公开的数据集分别进行特征提取与分类实验,实验结果表明,该算法在保持较高分类准确率的前提下,明显减少了特征数量,实现了较高的动态特征计算效率。

To improve the computational efficiency and performance of the feature selection of dynamic data streams,a feature selection algorithm of dynamic data streams based on rough set theory and artificial bee colony algorithm was proposed.The position update equations of employed bee phase and onlooker bee phase in the artificial bee colony algorithm were both modified to reduce the possibility of premature convergence of the artificial bee colony algorithm,the robustness of the artificial bee colony algorithm was enhanced,so that the modified artificial bee colony algorithm was suitable for dynamic feature selection algorithm.The fitness function of the incremental data was defined based on rough set,the new global feature set was abstracted from the old feature set and the incremental data using the artificial bee colony algorithm.Experimental results indicate that the proposed algorithm reduces the features significantly,and it realizes good dynamic feature computational efficiency,at the same time,it maintains high classification accuracy.