欢迎您！东篱公司
退出
-
申报数据库
-
立项数据库
-
成果数据库
-
项目获奖数据库
中文摘要：
针对球结构支持向量机（support vector machine，SVM）增量学习算法在训练时间和分类精度上的不足，提出了一种改进的球结构SVM多分类增量学习算法．该算法首先构造一个完全二叉树用于多类分类；分析新增样本的加入对原支持向量集的影响，将新增样本集中部分样本和原始训练集中的支持向量以及分布在球体一定范围内的样本合并做为新的训练集，完成分类器的重构．实现通过减少训练样本缩短训练时间和完善分类器提高分类精度的目的．通过UCI标准数据集实验，结果表明，该算法在所需训练的样本数、训练时间以及准确率3方面都优于球结构SVM增量学习算法，尤其当样本分布不平衡时，该算法有更高的分类准确率．
英文摘要：
Sphere-structured support vector machine （SVM） incremental learning algorithms perform poorly both in training time and classification precision. To solve this, an improved incremental multi-class classification learning algorithm based on a sphere-structured SVM was proposed. First, a full binary tree was constructed for multi-class classification. Next, possible changes to the support vector set were analyzed after new samples were added to the training set. Part of the samples in an incremental set, the support vectors in the original training set, and some samples within certain range of sphere were combined as a new training sample set to reconstruct the SVM classifier. This increased classification precision by reducing training samples, shortening training time and improving performance of the classifier. Using a UC Irvine （UCI） standard data set in an experiment, the results showed that the proposed algorithm is superior to the sphere-strnctured SVM incremental learning algorithm on the three aspects of training sample number, training time and classification precision. Especially in the case of unbalanced samples, the proposed algorithm demonstrated higher classification precision.
同期刊论文项目
同项目期刊论文
位置：