基于邻域容差熵选择集成分类算法  

Ensemble Classification Algorithm Selecting Based on Neighborhood-tolerance Entropy

作　　者：董红瑶 申成奥 李丽红 DONG Hongyao;SHEN Cheng′ao;LI Lihong(College of Science,North China University of Science and Technology,Tangshan 063210,China;Hebei Province Key Laboratory of Data Science and Application,Tangshan 063210,China;Tangshan Key Laboratory of Engineering Computing,Tangshan 063210,China)

机构地区：[1]华北理工大学理学院,河北唐山063210 [2]河北省数据科学与应用重点实验室,河北唐山063210 [3]唐山市工程计算重点实验室,河北唐山063210

出　　处：《郑州大学学报（理学版）》2023年第6期15-21,共7页Journal of Zhengzhou University:Natural Science Edition

基　　金：河北省数据科学与应用重点实验室项目(10120201);唐山市数据科学重点实验室项目(10120301)。

摘　　要：针对不完备混合型信息系统的分类问题,融合粒计算和集成学习思想,引入邻域容差关系,提出基于邻域容差熵选择集成分类算法。首先根据样本中的缺失属性将不完备混合型数据集划分为不同的信息粒,并再次遍历数据集进行最大化信息粒,构成新的粒空间,用以BP神经网络为基分类器的集成算法在粒空间上训练最大化信息粒,构建新的基分类器;然后以每个信息粒的缺失属性作为条件计算出关于类别属性的邻域容差条件熵,各个信息粒的重要度通过邻域容差条件熵进行量化后,通过信息粒的大小、新训练出的基分类器预测准确率以及邻域容差条件熵重新定义各个基分类器的权重;最后根据预测样本对基分类器加权集成,预测分类结果,并与传统的集成分类算法进行对比分析。对于不完备混合型数据集,新提出的集成分类算法能有效提升分类准确率。In order to solve the classification problem of incomplete hybrid information systems,a neighborhood-tolerance entropy selective ensemble classification algorithm based on neighborhood-tolerance relation was proposed by integrating the idea of granular computing and ensemble learning.Firstly,the incomplete mixed dataset was divided into different information grains according to the missing attributes in the samples,and the dataset was traversed again to maximize the grains to form a new grain space.In the grain space,the BP neural network was used as the basic classifiers to train the maximum information grains to construct a new basic classifier.And then with each information missing attributes as conditions the category attributes of the neighborhood-tolerance conditional entropy was caculated,the importance of information was quantified by neighborhood-tolerance entropy.By the size of the information grain,the prediction accuracy of the newly trained basic classifier and neighborhood-tolerance conditional entropy were used to redefine the weight of the basic classifiers.Finally,based on the prediction samples,the weighted ensemble of the basic classifier was used to predict the classification results.Compared with the traditional ensemble classification algorithms,for incomplete mixed datasets,the new ensemble classification algorithm could effectively improve the classification accuracy.

关 键 词：不完备混合型信息系统 信息粒 邻域容差熵 集成学习 分类 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]