CAJ | 학술논문

针对差异性是集成学习的必要条件,研究了基于k-means聚类技术提高神经网络分类器集成差异性的方法.通过训练集并使用神经网络分类器学习算法训练许多分类器模型,在验证集中利用每个分类器的分类结果作为聚类的数据对象;然后应用k-means聚类方法对这些数据聚类,在聚类结果的每个簇中选择一个分类器代表模型,以此构成集成学习的成员;最后应用投票方法实验研究了这种提高集成学习差异性方法的性能,并与常用的集成学习方法bagging、adaboost进行了比较.
침대차이성시집성학습적필요조건,연구료기우k-means취류기술제고신경망락분류기집성차이성적방법.통과훈련집병사용신경망락분류기학습산법훈련허다분류기모형,재험증집중이용매개분류기적분류결과작위취류적수거대상;연후응용k-means취류방법대저사수거취류,재취류결과적매개족중선택일개분류기대표모형,이차구성집성학습적성원;최후응용투표방법실험연구료저충제고집성학습차이성방법적성능,병여상용적집성학습방법bagging、adaboost진행료비교.
Aiming at diversity being a necessary condition of the ensemble learning,this paper studies the method for improving diversity of the neural networks ensemble based on k-means clustering technique.This paper proposes a selecting approach that is first to train many classifiers through training set with neural network algorithm,and uses the result by the classifiers from validation set for clustering.And then this paper uses the k-means algorithm to cluster the data set from the result and selects a classifier model from every cluster to make up of the membership of the ensemble learning.At last,this paper studies the performance of ensemble method by using vote method and compare performance with bagging and adaboost methods.