改进的超限学习机及其在不平衡数据中的应用  

An Improved Extreme Learning Machine and Its Application in Imbalanced Data

作　　者：李晗缦 王丽丹[1,2,3] 段书凯 LI Han-man;WANG Li-dan;DUAN Shu-kai(School of Electronic and Information Engineering,Southwest University,Chongqing 400715,China;National&Local Joint Engineering Laboratory of Intelligent Transmission and Control Technology,Chongqing 400715,China;Brain-Inspired Computing&Intelligent Control of Chongqing Key Lab,Chongqing 400715,China)

机构地区：[1]西南大学电子与信息工程学院,重庆400715 [2]智能传动与控制技术国家与地方联合工程实验室,重庆400715 [3]智能计算与智能控制重庆市重点实验室,重庆400715

出　　处：《西南大学学报（自然科学版）》2020年第6期140-148,共9页Journal of Southwest University(Natural Science Edition)

基　　金：国家自然科学基金项目(61571372,61672436);中央高校基本科研业务费专项资金项目(XDJK2016A001,XDJK2017A005).

摘　　要：超限学习机(ELM)作为一种简单高效的学习算法被广泛应用于分类和拟合问题中.但是ELM在训练过程中逼近所有的样本容易造成过拟合,并且单个的ELM在不平衡数据分类上效果欠佳.因此,本文提出了一种新的基于分层交叉验证的集成超限学习机,该算法在训练阶段将集成学习和分层交叉验证相结合:①集成学习通过将若干个网络组合大大提高分类性能;②分层交叉验证最大程度学习样本的分布特点.基于KEEL数据库的不平衡数据分类问题的实验表明,新提出的算法更加稳定并且有更高的分类性能.The extreme learning machine(ELM)has been widely applied in classification and regression learning due to its simple structure and fast learning capability.However,it might suffer from over-fitting as the network needs to approximate all samples,and single ELM is mediocre in imbalanced data classification.To deal with these problems,a novel ensemble extreme learning machine based on stratified cross-validation is proposed in this paper.Ensemble learning and stratified cross-validation are embedded into the training phase:①ensemble learning greatly improves classification performance by combining several basic networks;②stratified cross-validation could learn samples distribution characteristics to the greatest extent.Experimental results of imbalanced data sets show that the proposed method is robust and more efficient for imbalanced classification.

关 键 词：分层交叉验证 集成学习 超限学习机 不平衡数据分类 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]