Newton-Raphson算法Logistic分类器性能提升应用研究  

Study on the Application of Newton-Raphson Algorithm in Performance Improvement of Logistic Classifier

作　　者：叶晓波[1] 秦海菲 

机构地区：[1]楚雄师范学院网络与信息系统研究所,云南楚雄675000 [2]楚雄师范学院信息科学与技术学院,云南楚雄675000

出　　处：《软件导刊》2017年第11期141-143,148,共4页Software Guide

基　　金：云南省教育厅科学研究基金项目(2012Y131)

摘　　要：Logistic分类算法中,通常使用梯度上升算法求解权值矩阵,Newton-Raphson算法是另一种求解权值矩阵的算法。选用UCI机器学习仓库中3个数据集,实验研究了Logistic梯度上升算法与Logistic Newton-Raphson算法的分类正确率、权值矩阵迭代求解次数。实验结果表明,相比Logistic梯度上升算法,Logistic Newton-Raphson算法具有更高分类正确率与较少权值矩阵迭代求解次数,该结论为Logistic分类过程中权值矩阵求解算法提供了参考依据。The gradient ascent algorithm is usually used to solve the weight matrix in Logistic classifier,Newton-Raphson algorithm is another algorithm for solving the weight matrix.This paper selects three data sets in UCI machine learning repository,to do an experimental study on classification accuracy、iterative solution of weight matrix on Logistic gradient ascent algorithm and Logistic Newton-Raphson algorithm.The experiment results show thatLogistic Newton-Raphson algorithm has a higher classification accuracyand less iteration times for solution weight matrix,The results provide reference tosolve weight matrix for Logistic classifier.

关 键 词：模式识别 LOGISTIC NEWTON-RAPHSON 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]