【摘要】 为了解决核爆地震自动识别中最优特征子集的选择问题,根据克隆选择原理,提出了一种过滤与封装相结合的特征选择方法.该方法融合了封装式与过滤式特征选择方法的优点,利用局部化的类别可分性判据来处理核爆地震样本的多峰分布问题,通过设定独立的记忆抗体能够保证最终结果是搜索过的最佳特征组合,并且可以处理设定和不设定最优特征子集维数两种情况下的特征选择问题.首先通过UCI数据集中呈多峰分布的玻璃数据验证了该特征选择方法的有效性,进而将其应用到核爆地震特征选择中.核爆地震特征选择实验结果表明,该方法不仅有效地降低了特征空间的维数,而且使分类精度提高了2个百分点,与封装式特征选择方法相比,该方法的计算复杂度大为降低.更多还原

【Abstract】 A filter-wrapper feature selection algorithm is proposed based on the clonal selection principle to select the optimal feature subset in the automatic classification of nuclear explosions and earthquakes.This algorithm possesses the merits of both filter and wrapper.It utilizes local class separability criterion to deal with multimodal data of nuclear explosions and earthquakes.A separate memory antibody is set which can guarantee the output is the best feature combination among that was searched and can deal with both cases under which the optimal dimensionality is specified in advance or not.Firstly,the effectiveness is validated using glasses data of UCI whose distribution is multimodal,and then it is applied to feature selection of nuclear explosions and earthquakes.The experimental results on seismic data show that the proposed algorithm reduces feature dimensionality effectively while improves the recognition accuracy by 2 percent. Moreover,the computation complexity is less than that of wrapper.更多还原