为改进基于数据描述的单类分类机识别率,将样本分布密度加入分类机的设计中,提出采用密度诱导型数据描述单类分类机(DISVDD).以支撑向量域描述(SVDD)算法为基础,通过一种简易的形式引入数据密度因子,使高密度区数据对分类支撑域的作用被强化,而低密度区数据的作用被削弱,结果使分类超球体因靠近高密区而提高其识别性能,而且不增加计算复杂度.在构造样本值与真实数据集上的实验结果表明,所提出的算法对于不同类型的数据均具有更好的推广性.
The density induced support vector data description(DISVDD) is proposed to improve the recognition rate of one-class classifier by introducing the structure of the given data.Based on SVDD,DISVDD ensures that the learn ability of data domain having higher density structure is enhanced.Meanwhile,the learn ability of data domain having lower density structure is weakened.Consequently,DISVDD has better domain margin and gets higher recognition rate without the increase of computation load.Experiments with various real datasets show better promising results.