Hadoop平台分布式SVM算法分类研究

摘要

随着大数据的发展,分布式支持向量机(SVM)成为该领域研究热点.传统层级分布式SVM算法(CascadeSVM),在Hadoop平台下寻找全局最优支持向量的过程十分缓慢.本文提出了一种改进方法,先将传统的网格法与粒子群(PSO)算法结合,改进了单机PSO算法,再将单机PSO算法与Hadoop平台结合实现了一种新型卫星并行PSO算法(NPP-PSO).实验结果表明,相比于单机SVM算法,本文的分布式SVM算法,在保证了准确率的前提下大幅提高了计算速度;而使用NPP-PSO参数寻优后的分布式SVM,分类准确率相比于分布式SVM算法又有了明显提高.%With the development of big data,distributed support vector machine (SVM) has become a hot research topic in this field.The process of finding the global optimal support vector in the Hadoop platform is long under the traditional hierarchical Cascade SVM algorithm.This paper presents an improved method by firstly combining the traditional grid method and the particle swarm optimization(PSO) algorithm to improve the PSO algorithm.And a new satellite parallel PSO algorithm is realized by combining the single machine PSO algorithm and the Hadoop platform (NPP-PSO).The experimental results show that compared with the single SVM algorithm,the distributed SVM algorithm cannot only ensure the accuracy but can also greatly boost the computation speed.With the wide use of NPP-PSO distributed SVM,the classification accuracy has improved significantly.