一种基于关联规则的核粒度支持向量机

A Kernel Granular Support Vector Machine Based on Association Rules

提出一种基于关联规则的核粒度支持向量机(association rules based kernel granular SVM,AR-KGSVM)学习算法。AR-KGSVM首先将输入空间中的样本用核函数映射到高维特征空间,然后在核特征空间挖掘基于距离度量的关联规则以划分粒。算法的粒划分和数据训练都是在高维核空间中进行,避免了一般的粒度支持向量机(granular SVM,GSVM)在低维空间作粒划分而在高维空间中训练,使数据分布不一致而导致泛化能力不高的问题。在标准数据集上的实验结果表明AR-KGSVM的泛化能力优于传统的SVM和GSVM方法。

This paper proposes a kernel granular support vector machine based on association rules（AR- KGSVM）learning approach. Firstly, samples in input space are mapped into high-dimensional feature space and then association rules based on distance metric are mined in kernel future space. These association rules are applied to divide granules. Both the granular dividing and data training are executed in the high-dimensional kernel space. Hence,the proposed AR-KGSVM can avoid the poor generalization performance like traditional granular SVM （GSVM） due to the inconsistent data distribution,where granules are divided in the low-dimensional space but samples are trained in high-dimensional space. The ex- perimental results on benchmark datasets demonstrate that the performance of AR-KGSVM is superior to that of traditional SVM and GSVM.