论文部分内容阅读
为解决小样本回归时引起的过学习问题并提高回归精度,提出一种基于非线性空间特征选择的支持向量机.该方法依据矩阵相似度量或从研究的实际问题出发,绕过核技巧,直接将原始输入空间映射为适宜的非线性空间.该方法运用遗传算法在维数较多的非线性空间中搜索对输出影响最大的一些特征,达到降低输入空间维数的目的,从而避免过学习问题,并可获得简明的非线性回归函数.
In order to solve the over-learning problem caused by regression of small samples and improve the regression accuracy, a support vector machine (SVM) based on nonlinear spatial feature selection is proposed, which is based on the similarity measure of the matrix or from the actual problem of the study. The original input space is mapped to a suitable non-linear space.This method uses genetic algorithm to search some features that have the most influence on output in non-linear space with many dimensions, so as to reduce the dimension of input space and avoid over-learning problem , And can be concise nonlinear regression function.