摘要
针对支持向量回归因时空复杂度较高而无法处理大规模数据的问题,提出了一个新颖的增量学习模型——L增量υ支持向量回归(L IncrementalυSupport Vector Regression,LISVR)。该模型针对支持向量丢失所产生的不利影响,通过不断对支持向量样本加权并及时淘汰非支持向量,降低了时空复杂度。从理论上证明了算法可收敛到全局最优解。结合人工数据集、UCI数据集和机场噪声的实际问题对该算法做了相应测试,结果验证了算法的有效性。
In view of costly time-space complexity for support vector regression in dealing with large scale data,this paper presented a novel algorithm named the L Incremental v Support Vector Regression (LISVR) by means of incremental learning.LISVR eliminates non-support vectors each iteration and then takes the support vectors as the training samples with the weight factor.It reduces time-space complexity and enhances the regression results simultaneously.Theoretically,this paper proved the convergence of the global optimal solution.The experiments on the artificial data sets,UCI data set and airport noises show the effectiveness of the LISVR.
出处
《计算机科学》
CSCD
北大核心
2014年第6期166-170,共5页
Computer Science
基金
国家自然科学基金(61139002)资助
关键词
支持向量回归
支持向量
增量学习
机场噪声
Support vector regression
Support vectors
Incremental learning
Airport noise