基于小样本SVR的迁移学习及其应用

当前机器学习的技术已经运用到很多工程项目中,但大部分机器学习的算法只有在样本数量充足且运用在单一场景中的时候,才能获得良好的结果。其中,经典的支持向量回归机是一种具有良好泛化能力的回归算法。但若当前场景的样本数量较少时,则得到的回归模型泛化能力较差。针对此问题,以加权ε支持向量回归机为基础,提出了一种小样本数据的迁移学习支持向量回归机算法。该算法以加权ε支持向量回归机为Bagging算法的基学习器,使用与目标任务相关联的源域数据,通过自助采样生成多个子回归模型,采用简单平均法合成一个总回归模型。在UCI数据集和现实数据集——玉米棒与花生粒储藏环节损失数据集上的实验结果表明,该算法较标准ε-SVR算法与改进的RMTL算法在小数据样本上有更好的泛化能力。

Transfer Learning Based on Support Vector Regression Model for Small Sample Data and Its Applications

Machine learning technologies have been applied to many industry programs nowadays,but most of them can obtain satisfied results with sufficient samples in a single situation.For instance,the classical support vector regression is a regression algorithm with better generalization ability.However,if the sample size in the current scene is small,the generalization ability of the regression model is poor.To solve this problem,we propose a transfer learning support vector regression algorithm for small sample data based on weighted ε support vector regression.In this paper,ε weighted support vector regression is taken as the basic learner of Bagging algorithm,and multiple sub regression models are generated by bootstrap using source data associated with target data,and a general regression model is synthesized by simple average method.Experimental results on the UCI datasets and the real dataset,the corn and peanut sales loss dataset,show that the proposed algorithm has better generalization ability than SVR algorithm and the improved RMTL algorithm on small data samples.