论文部分内容阅读
为解决传统支持向量机易出现学习“过拟合”和丢失数据统计特征等问题,通过引入模糊隶属度和总间隔思想,提出一种基于总间隔的最大间隔最小包含模糊球形学习机(TMF-SSLM),使得一类(正类)被包含于一个最小包含超球内,而另一类(负类)与该超球间隔最大化,从而同时实现类间间隔的增大和正负两类类内体积的缩小.通过使用差异成本,解决不平衡训练样本问题.引入总间隔和模糊性惩罚,克服传统软间隔分类机的过拟合问题,显著提升球形学习机的泛化能力.采用UCI实际数据集分别对二类和一类模式分类进行实验,结果显示TMF-SSLM具有优于相关方法的稳定分类性能.
In order to solve the problems of traditional support vector machine (SVM), such as easy to learn, overfitting and statistical data loss, the fuzzy interval learning algorithm based on total interval is introduced. TMF-SSLM) such that one class (positive class) is contained within one minimal inclusion hypersphere while the other class (negative class) maximizes the separation from the hypersphere, thereby achieving both an increase in class spacing and plus or minus two The volume of the class is reduced.Using the difference cost to solve the problem of unbalanced training samples.This paper introduces the general interval and fuzziness penalty to overcome the over-fitting problem of the traditional soft-interval classifier and significantly improve the generalization ability of spherical learning machine.Using UCI actual dataset experimentally tests two classes and one class respectively. The results show that the TMF-SSLM has stable classification performance over the related methods.