【摘要】 核向量机可以高效学习大样本数据集,却有泛化能力低的缺陷.针对已有参数C选择算法缺乏启发性以及选取困难的不足,本文在分析了核聚类算法和距离比较算法的基础之上,提出基于核聚类的相对距离比较方法,该算法利用核聚类算法在特征空间对样本点进行聚类分簇,然后根据样本点到簇心相对距离的比值,得到参数C.本文在理论和实验两个方面,证明该算法有效地选择参数C,从而提高核支持向量机算法的泛化能力.更多还原

【Abstract】 Core vector machine can learn the large dataset efficiently,but has Low generalization ability.The selection algorithms of parameter C are not heuristic and difficult.Based on the analysis of kernel clustering algorithm and distance comparison algorithm,this paper proposes kernel clustering based on the relative distance comparison method.The algorithm divides the sample points in the feature space into clusters by kenel clustering algorithm.then it gets C parameter on the ratio of relative distance of cluster center.In this paper,the algorithm is proved to choose parameter C efficiently and enhance generalization ability of core vector machine algorithm in theory and experiment.更多还原