摘要： 数据描述只使用目标集训练样本获得关于目标集的描述,支持向量数据描述(SVDD)是一种有效的数据描述方法。样本错误加权的SVDD(WSVDD)推广了SVDD,对每个训练样本的错误赋予不同的权值,可以精细地控制训练样本对超球面边界的影响。用UCI机器学习数据集的两个数据和图标分类的实验验证了WSVDD的有效性。

关键词: 机器学习, 数据描述, 支持向量数据描述, 样本错误加权的支持向量数据描述

Abstract: By employing training examples of target set only, data description is able to obtain description of target set, and SVDD(support vector data description) is an efficient data description scheme. In this paper SVDD is generalized to propose WSVDD(example error weighted support vector data description). By giving various weight , how each training example affect the boundary of hypersphere could be controlled . Two test data from UCI machine learning repository are employed to evaluate the usefulness of WSVDD, as well as logo classification task.

中图分类号: 

• TP18