基于MapReduce的并行增量迭代支持向量机算法  

PARALLELISED INCREMENTAL ITERATIVE SUPPORT VECTOR MACHINE BASED ON MAPREDUCE

作　　者：马莹莹[1] 王黎明[1] 王世卿[1] 

机构地区：[1]郑州大学信息工程学院,河南郑州450000

出　　处：《计算机应用与软件》2015年第4期288-291,共4页Computer Applications and Software

摘　　要：为了提高支持向量机处理大数据集的性能,在研究MapReduce编程模型的基础上结合增量学习及迭代技术提出了改进算法MRII-SVM。该算法采用增量学习方式,在Map阶段并行求解各数据切片的支持向量集,并利用KKT条件削减增量数据集,在Reduce阶段执行联合重训练使算法收敛于当前最优解。基于UCI标准数据集的实验结果显示,MRII-SVC在保持SVM算法分类精度的基础上,提高了数据处理能力和运行效率。In order to improve the performance of the support vector machine in processing large-scale datasets,we propose an improved SVM algorithm,named MRII-SVM,based on studying MapReduce programming model and combining incremental learning and iterative technology.The algorithm adopts incremental learning means,calculates the support vector set of each data slice parallelly in Map phase,and uses KKT condition to curtail incremental dataset,executes combined retraining in Reduce phase to make the algorithm converge to current optimal solution.Experiment results on UCI standard data sets show that MRII-SVM guarantees the high accuracy of the standard SVM classifierand simultaneously improves the capability and efficiency of data processing.

关 键 词：支持向量机 并行 增量 迭代 MAPREDUCE 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]