基于改进引力搜索算法的优化特征选择算法

Optimized feature subset selection based on improved gravitational search algorithm

针对引力搜索算法早熟收敛和局部收敛能力慢的不足,提出一种改进的引力搜索算法(BGSA-PS),并用于特征选择处理,从原始特征集合中寻找合适且数量较小的特征子集。加入多样性因子更新粒子的速度,扩展全局搜索空间,防止早熟收敛,结合模式搜索法增强并加速局部搜索能力。在UCI分类数据集上的实验结果表明,该方法同原始离散型引力搜索算法及相似算法相比,选取的特征数量较少、分类精度较高,是一种有效的特征选择方法,可广泛用于特征选择领域。

To overcome the premature search convergence and the stagnation situation,an algorithm based on improved binary gravitational search algorithm（BGSA-PS）was proposed and applied for feature subset selection,which selected the best subset of features.The diversity factor was introduced into BGSA,which extended the search space to prevent the premature search convergence.The pattern search method was combined to increase the local search capacity.Experimental results on UCI datasets show the effectiveness and efficiency of the proposed algorithm.Smaller number of selected features is obtained and higher classification accuracy is achieved than using BGSA and similar algorithms.The algorithm can be successfully applied in the field of feature selection.