不平衡数据加权边界点集成欠采样方法

doi:10.19665/j.issn1001-2400.2021.04.023

摘要/Abstract

摘要：

为了有效地解决不平衡数据中边界点直接被删除的问题,保持多数类数据的信息,提出一种基于聚类的加权边界点集成欠采样算法。首先该算法通过提取少数类数据集的数据点作为多数类数据集的初始聚类中心个数进行聚类;然后引入变异系数将边界点识别出来,对识别出的边界点进行加权,使得加权后的边界点可以加入到不平衡数据的处理中;再利用簇密度将多数类数据集分为高密度簇和低密度簇,把低密度簇删除;最后获得约简后的多数类样本集。再将约简后的多数类样本和少数类样本结合,形成平衡的数据集,利用Ada Boost对其进行训练,得到最终的分类模型。利用该方法可以实现对数据集的约简操作,提高执行效率。结果表明,所提方法可以有效地处理不平衡数据的问题,提高了不平衡数据加权边界点集成欠采样算法的执行效率和结果的精确性。

关键词: 采样, 聚类, 不平衡数据, 加权边界点

Abstract:

In order to effectively solve the problem that boundary points are deleted directly from unbalanced data and effectively maintain the information on most kinds of data,a clustering-based weighted boundary point integration undersampling algorithm is proposed.First,the algorithm extracts the number of minority class sets as the initial number of clustering centers of majority class sets to cluster.Then,the variation coefficient is introduced to identify the boundary points,and the identified boundary points are weighted so that the weighted boundary points can be added to the unbalanced data processing.Then,the cluster density is used to divide majority class sets into the high-density cluster and low-density cluster,delete the low-density cluster,and finally obtain the reduced majority of the sample sets.Then,the reduced majority of class samples is combined with the minority of class samples to form a balanced data set,which is trained with the Ada boost to get the final classification model.This method can be used to reduce the dataset and improve the efficiency of execution.The results show that the proposed method can effectively handle the problem of unbalanced data,and improve the execution efficiency and accuracy of the under-sampling algorithm for unbalanced data weighted boundary point integration.

Key words: sampling, clustering, unbalanced data, weighted boundary point

中图分类号:

TP311.13

何云斌,冷欣,万静. 不平衡数据加权边界点集成欠采样方法[J]. 西安电子科技大学学报, 2021, 48(4): 176-183.

HE Yunbin,LENG Xin,WAN Jing. Unbalanced data weighted boundary point integration undersampling method[J]. Journal of Xidian University, 2021, 48(4): 176-183.

图/表 6

图1

图2

表1

表2

表3

图3

参考文献 16

[1]	陈旭, 刘鹏鹤, 孙毓忠, 等. 面向不均衡医学数据集的疾病预测模型研究[J]. 计算机学报, 2019, 42(03):596-609.
	CHEN Xu, LIU Penghe, SUN Yuzhong, et al. Research on Disease Prediction Models Based on Imbalanced Medical Data Sets[J]. Chinese Journal of Computers, 2019, 42(03):596-609.
[2]	FIOREU, DE S A, PERLA F, et al. Using Generative Adversarial Networks for Improving Classification Effectiveness in Credit Card Fraud Detection[J]. Information Sciences, 2017, 479:448-455. doi: 10.1016/j.ins.2017.12.030
[3]	宋胜利, 王少龙, 陈平. 面向文本分类的中文文本语义表示方法[J]. 西安电子科技大学学报, 2013, 40(02):89-97.
	SONG Shengli, WANG Shaolong, CHEN Ping. Semantic Representation of Chinese Text for Text Classification[J]. Journal of Xidian University, 2013, 40(02):89-97.
[4]	AIHONG W, NAN Y, CAO X. Multi-Classification Cluster Analysis of Large Data Based on Knowledge Element in Microblogging Short Text[J]. Cluster Computing, 2019, 22(2):4119-4127. doi: 10.1007/s10586-017-1517-9
[5]	GUPTA M, BAKLIWAL A, AGARWAL S, et al. A Comparative Study of Spam SMS Detection Using Machine Learning Classifiers[C]//Proceedings of the 2018 Eleventh International Conference on Contemporary Computing (IC3).Piscataway:IEEE, 2018:1-7.
[6]	MARTIN-DIAZ I, MORINIGO-SOTELO D, DUQUE-PEREZ O, et al. Early Fault Detection in Induction Motors Using AdaBoost with Imbalanced Small Data and Optimized Sampling[J]. IEEE Transactions on Industry Applications, 2017, 53(3):3066-3075. doi: 10.1109/TIA.2016.2618756
[7]	LIN W C, TSAI C F, HU Y H, et al. Clustering-Based Undersampling in Class-Imbalanced Data[J]. Information Sciences, 2017, 409:17-26.
[8]	熊冰妍, 王国胤, 邓维斌. 基于样本权重的不平衡数据欠抽样方法[J]. 计算机研究与发展, 2016, 53(11):2613-2622.
	XIONG Bingyan, WANG Guoyin, DENG Weibin. Under-Sampling Method Based on Sample Weight for Inbalanced Data[J]. Journal of Computer Research and Development, 2016, 53(11):2613-2622.
[9]	KANG Q, SHI L, ZHOU M C, et al. A Distance-Based WeightedUndersampling Scheme for Support Vector Machines and Its Application to Imbalanced Classification[J]. IEEE Transactions on Neural Networks, 2018, 29(9):4152-4165.
[10]	蔡艳艳, 宋晓东. 针对非平衡数据分类的新型模糊SVM模型[J]. 西安电子科技大学学报, 2015, 42(05):120-124.
	CAI Yanyan, SONG Xiaodong. New Fuzzy SVM Model Used in Imbalanced Datasets[J]. Journal of Xidian University, 2015, 42(05):120-124.
[11]	李钊, 袁文浩, 任崇广, 等. 跨层精度自动调节的k均值聚类近似计算方法[J]. 西安电子科技大学学报, 2020, 47(03):50-57.
	LI Zhao, YUAN Wenhao, REN Chongguang, et al. Approximate Computing Method Based on Cross-Layer Dynamic Precision Scaling For the k-means[J]. Journal of Xidian University, 2020, 47(03):50-57.
[12]	薛丽香, 邱保志. 基于变异系数的边界点检测算法[J]. 模式识别与人工智能, 2009, 22(5):799-802.
	XUE Lixiang, QIU Baozhi. Boundary Points Detection Algorithm Based on Coefficient Variation[J]. Pattern Recognition and Artificial Intelligence, 2009, 22(5):799-802.
[13]	BREUNIG M M, KRIEGEL H P, NG R T. LOF:Identifying Density-Based Local Outliers[C/OL].[2020-04-30].https://www.docin.com/P-1572662359.html .
[14]	XIA C, HSU W, LEE M L, et al. BORDER:Efficient Computation of Boundary Points[J]. IEEE Transactions on Knowledge and Engineering, 2006, 18(3):289-303.
[15]	黄浩, 何钦铭, 陈奇. 基于加权边界度的稀有类检测算法[J]. 软件学报, 2012, 23(05):1195-1206.
	HUANG Hao, HE Qinming, CHEN Qi. Rare Categeory Detection Algorithm Based on Weighted Boundary Dedree[J]. Jouanal of Software, 2012, 23(05):1195-1206.
[16]	FREUND Y, SCHAPIRE R E. A Decision-Theoretic Generalization of On-Line Learning and An Application to Boosting[J]. Journal of Computer and System Sciences, 1997, 55(1):119-139. doi: 10.1006/jcss.1997.1504

数据集	样本数量	少数类数据集	多数类数据集	不平衡率/%
Glass	214	29	185	6.38
Vehicle	846	199	647	3.25
Satimage	6 435	626	5 809	9.28
Vowel	990	90	900	10.00
Pima	768	268	500	1.87
Haberman	306	81	225	2.78
Letter	20 000	734	19 266	26.25

数据集	C4.5	SMOTE-Boost	PC-Boost	Ada-Boost	CWBUSC
Glass	0.785	0.846	0.978	0.813	0.981
Vehicle	0.879	0.617	0.862	0.554	0.901
Satimage	0.564	0.678	0.657	0.647	0.784
Vowel	0.937	0.955	0.975	0.971	0.991
Pima	0.600	0.648	0.519	0.681	0.702
Haberman	0.323	0.391	0.407	0.369	0.423
Letter	0.947	0.986	0.969	0.691	0.951

数据集	C4.5	SMOTE-Boost	PC-Boost	Ada-Boost	CWBUSC
Glass	0.859	0.911	0.949	0.894	0.950
Vehicle	0.925	0.953	0.962	0.955	0.977
Satimage	0.727	0.760	0.822	0.770	0.853
Vowel	0.958	0.987	0.921	0.976	0.989
Pima	0.661	0.724	0.767	0.703	0.789
Haberman	0.486	0.556	0.674	0.523	0.680
Letter	0.937	0.975	0.958	0.848	0.948