论文部分内容阅读
为了得到一个低误分类代价的特征子集,本文通过定义样本间的代价距离并将代价距离引入了现有的特征选择架构,把流形学习和代价敏感特征选择问题相结合得到了一个新的代价敏感特征选择方法,称之为基于流形学习的代价敏感特征选择算法。以前提出的代价敏感特征选择算法在选择特征的过程中只考虑到了特征与误分类代价的关系,并对特征一个一个的进行选择,而本文所提出的代价敏感特征选择算法同时考虑了特征与误分类代价的关系和特征之间内在的判别信息,从而提高了代价敏感特征选择效果。在六个现实世界数据集上的实验证明了本文所提出的算法效果优于现有的相关算法。
In order to obtain a subset of features with low misclassification cost, this paper introduces a cost-aware feature selection problem by defining the cost distance between samples and introducing the cost distance into the existing feature selection framework. The cost-sensitive feature selection method is called a cost-sensitive feature selection algorithm based on manifold learning. The previously proposed cost-sensitive feature selection algorithm only considers the relationship between feature and misclassification cost and selects the features one by one. The cost-sensitive feature selection algorithm proposed in this paper considers both the feature and the error Classification of the relationship between the cost and characteristics of the inherent discriminant information, thereby increasing the cost-sensitive feature selection effect. Experiments on six real-world datasets prove that the proposed algorithm is better than the existing one.