基于RUSBoost和积矩系数的神经网络优化算法  

Optimization algorithm for neural network based on RUSBoost and correlation coefficient

作　　者：尹化荣 陈莉[1] 张永新[1] 陈丹丹[1] Yin Huarong;Chen Li;Zhang Yongxin;Chen Dandan(School of Information Science&Technology,Northwest University,Xi’an 710127,China)

机构地区：[1]西北大学信息科学与技术学院,西安710127

出　　处：《计算机应用研究》2018年第9期2592-2596,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(61502219);博士后基金资助项目(2015M582697)

摘　　要：针对单个神经网络分类准确率低、RUSBoost算法提高NN分类器准确率耗时长的问题,提出了一种混合RUSBoost算法和积矩系数的分类优化算法。首先,利用RUSBoost算法生成m组训练集;然后,依据Pearson积矩系数计算每组训练集属性的相关程度消除冗余属性,生成目标训练集;最后,新的子训练集训练神经网络分类器,选择最大准确率分类器作为最终的分类模型。实验中使用了四个Benchmark数据集来验证所提算法的有效性。实验结果表明,提出的算法的准确率相较于传统的算法最多提升了8.26%,训练时间最高降低了62.27%。Referring to the current problems that a single neural network classification accuracy is low,and RUSBoost is a time-consuming algorithm for improving the NN classifier accuracy.This paper proposed a hybrid algorithm which combined RUSBoost with the Pearson correlation coefficient to optimize the classifier of neural network.First of all,it generated m by using RUSBoost algorithm for group training.Then,according to the Pearson product-moment coefficient of correlation calcula-ted the property of each group and eliminated redundant properties in the set,generating target training sets.Finally,it trained neural network classifiers by the new training set,and selected the maximum accuracy classifier as the ultimate model.It chose 4 Benchmark data sets to verify the effectiveness of the algorithm.Experimental results show that the accuracy of the algorithms presented compared to traditional algorithm for maximum lifting 8.26%,and the training timer reduced 62.27%.

关 键 词：神经网络 RUSBoost 积矩系数 集成学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]