摘要
对于拟合问题,传统的模式识别特征筛选方法以各特征量对训练数据拟合能力的贡献为取舍标准,未考虑经验风险最小化和结构风险最小化间的差别,不能获得预报能力最强的特征筛选结果。为此我们提出了结合支持向量回归法与留一法的特征筛选新算法,并将它试用于镍氢电池材料和氧化铝净溶出率两套实验数据集的特征筛选。
Most of the traditional feature selection methods designed for regression problem only consider the consistence between the training data and the regression result. However, they neglect the vital differences betvveen the empirical risk minimization and the structural risk minimization and hence cannot directly find the feature subset with high generalization performance. To solve this problem, this paper proposed a floating search method for feature selection, which was based on support vector regression and leaving-one method. Experiments were carried out on two chemical data sets.
出处
《计算机与应用化学》
CAS
CSCD
北大核心
2002年第6期703-705,共3页
Computers and Applied Chemistry
基金
由国家自然科学基金和上海宝钢集团公司联合资助(50174038)
