基于秩空间差异的多核组合方法

在求解秩空间差异性的基础上,提出一种基于秩空间差异性的多核组合方法。将样本按照特征进行分组,使用不同的核函数对已完成分组的数据进行训练,并应用网格搜索法对核函数的参数进行寻优。在备选核函数中选取两个核函数,将分为两组的数据分别放入对应的核函数中进行映射,通过判断数据经过核函数映射后的秩空间差异性为基础核函数的选择提供参考。选用白酒数据集、乳腺癌数据集以及葡萄酒品质数据集进行试验,验证了当数据经过已选的基础核函数映射后,秩空间差异越大,分类的准确率越高。试验结果表明应用该方法进行基础核函数的选择以及组合的可行性。

Multi-kernel combination method based on rank spatial difference

A multi-kernel combination method based on rank spatial difference was proposed in this paper. samples were grouped according to characteristics, different kernel functions are used to train the grouped data, and the parameters of the kernel function are optimized by grid search method. Two kernel functions were selected from the alternative kernel functions, and the data divided into two groups were respectively put into the corresponding kernel function for mapping. Then the rank spatial difference of the data after the kernel function mapping was judged to provide reference for the selection of the basic kernel function. The wine data set, the breast cancer data set and the wine quality data set were selected for the experiment to verify that when the data were mapped by the selected basic kernel function, the greater the rank space difference was, the higher the classification accuracy was. The experimental results showed that the method was feasible for the selection and combination of basic kernel functions.