论文部分内容阅读
针对传统支持向量机不能较好地利用数据空间局部信息的问题,提出一种基于局部学习的支持向量机.通过同时最小化局部内散度和最大化局部间散度信息来寻求一个最优的分类决策函数.为了更好地反映数据的局部几何特征,该方法采用适于局部学习的测地线距离来度量数据点对间的相似性.另外,通过引入一个能同时控制间隔误差上界和支持向量下界的参数,进一步提升学习泛化能力.人造和实际数据集实验验证了所提出方法的有效性.
Aiming at the problem that traditional support vector machines can not make good use of the local information in data space, this paper proposes a support vector machine based on local learning to find out the best one by minimizing the local internal divergence and maximizing the local divergence information In order to better reflect the local geometric features of the data, this method uses the geodesic distance suitable for local learning to measure the similarity between data point pairs.In addition, by introducing an algorithm that can control the interval error upper bound and Support vector lower bound parameters to further enhance the learning generalization ability.The artificial and actual data sets experimentally verify the effectiveness of the proposed method.