一种基于MA-LSSVM的封装式特征选择算法

Wrapper feature selection algorithm based on MA-LSSVM

为了解决高维小样本的特征选择问题,该文结合文化基因算法(Memetic algorithm,MA)与最小二乘支持向量机(Memetic algorithm and least squares support vector machine,MALSSVM),设计了一种封装式(Wrapper)特征选择算法。该方法将全局搜索与局部搜索相结合作为求解策略,利用了最小二乘支持向量机易于求解的特点,构造分类器,以分类的准确率作为文化基因算法寻优过程中适应度函数的主要成分。实验表明,MA-LSSVM可以较高效稳定地获取对分类贡献较大的特征,降低数据维度,提高了分类效率。

To improve the feature selection problem of the high dimensional small sample data, this paper combines memetic algorithm （MA）and least squares support vector machine （LS-SVM）to design a wrapper feature selection method （MA-LSSVM）. The solving strategy of the proposed method is composed by global search and local search, which utilizes the speciality of being easy to search optimal solution to construct classifiers and to regard classification accuracy as the main component of memetic algorithm fitness function in the optimization process. The experimental results demonstrate that the MA-LSSVM can be more efficient and stable to obtain features larger contribution to the classification precision, reducing the data dimension and improving the classification efficiency.