基于条件概率分布的混合距离度量方法及应用

doi:10.3969/j.issn.1000-1565.2025.05.008

河北大学学报(自然科学版) ›› 2025, Vol. 45 ›› Issue (5): 520-529.DOI: 10.3969/j.issn.1000-1565.2025.05.008

基于条件概率分布的混合距离度量方法及应用

胡桂开,杨沛融

收稿日期:2024-10-15 发布日期:2025-09-18
作者简介:胡桂开(1977—),男,东华理工大学教授,博士,主要从事机器学习及其应用和回归模型统计推断的研究.
E-mail:hgk1997@163.com
基金资助:
国家自然科学基金项目(11661003);江西省自然科学基金项目(20192BAB201006)

Hybrid distance measurement method and application based on conditional probability distribution

HU Guikai, YANG Peirong

School of Science, East China University of Technology, Nanchang 330013, China

Received:2024-10-15 Published:2025-09-18

摘要/Abstract

摘要： 为提高名词性属性实例差异的识别精度,优化分类算法性能,综合考虑实例的属性和类别特征,提出了一种基于条件概率分布的混合距离度量方法.首先,计算属性间以及属性与类别间条件概率分布的差异;其次,利用互信息对2种差异进行加权组合,得到新的混合距离度量;最后,利用K-近邻算法在20个UCI(University of California Irvine)数据集上进行仿真实验,并将其应用于儿童阑尾炎的诊断和治疗.结果表明:较重叠度量等3种度量方法,本文提出的距离度量方法,显著提高了分类算法的准确率,具有较好的应用前景.

关键词: 条件概率分布, 混合距离度量, 互信息, 名词性属性

Abstract: To improve the recognition accuracy of nominal attribute instance differences and optimize the performance of classification algorithms, a hybrid distance measurement method based on conditional probability distribution is proposed, which comprehensively considers the attributes and category features of instances.Firstly, this method calculates the differences in conditional probability distributions among attributes and between attributes and categories. Secondly, a new hybrid distance metric is obtained by using mutual information to weight and combine the two differences. Finally, simulation experiments are performed on 20 UCI datasets based on K-nearest neighbor algorithm. Meanwhile, the distance measurement method is applied to the diagnosis and treatment of appendicitis in children. The results show that compared with the three measurement methods including overlap measurement, the distance measurement method proposed in this work significantly improves the accuracy of the classification algorithm and has good application prospects.

Key words: conditional probability distribution, hybrid distance metric, mutual information, nominal attribute

中图分类号:

TP301.6

胡桂开,杨沛融. 基于条件概率分布的混合距离度量方法及应用[J]. 河北大学学报(自然科学版), 2025, 45(5): 520-529.

HU Guikai, YANG Peirong. Hybrid distance measurement method and application based on conditional probability distribution[J]. Journal of Hebei University(Natural Science Edition), 2025, 45(5): 520-529.

参考文献

[1] DOMINGOS P. Rule induction and instance-based learning: A unified approach[C] //International Joint Conference on Artificial Intelligence,Morgan Kaufmann Publishers Inc, 1995:1226-1232.
[2] LI C Q, JIANG L X, LI H W. Naive Bayes for value difference metric[J]. Front Comput Sci, 2014, 8(2):255-264. DOI: 10.1007/s11704-014-3038-5.
[3] MASTELINI S M, DE LEON FERREIRA DE CARVALHO A C P. Efficient online tree, rule-based, and distance-based algorithms[C] //Anais do XXXVII Concurso de Teses e Dissertações(CTD 2024), Brasil, Sociedade Brasileira de Computação-SBC, 2024: 21-27. DOI: 10.5753/ctd.2024.1859.
[4] AHA D W, KIBLER D, ALBERT M K. Instance-based learning algorithms[J]. Mach Learn, 1991, 6(1): 37-66. DOI: 10.1007/BF00153759.
[5] BLANZIERI E, RICCI F. Probability based metrics for nearest neighbor classification and case-based reasoning[M] //Lecture Notes in Computer Science, Berlin, Heidelberg: Springer Berlin Heidelberg, 1999: 14-28. DOI: 10.1007/3-540-48508-2_2.
[6] QUANG L S, BAO H T. A conditional probability distribution-based dissimilarity measure for categorial data[C] //DAI H, SRIKANT R, ZHANG C. Pacific-Asia Conference on Knowledge Discovery and Data Mining, Berlin, Heidelberg: Springer, 2004: 580-589. DOI: 10.1007/978-3-540-24775-3_69.
[7] DE SANTIS E, MASCIOLI F M F, SADEGHIAN A, et al. A dissimilarity learning approach by evolutionary computation for faults recognition in smart grids[C] //International Joint Conference on Computational Intelligence. Cham: Springer, 2016: 113-130. DOI: 10.1007/978-3-319-26393-9_8.
[8] 李仁侃,叶东毅.粗糙K-Modes聚类算法[J].计算机应用, 2011, 31(1): 97-100. DOI: 10.3724/SP.J.1087.2011.00097.
[9] 李超群.名词性属性距离度量问题及其应用研究[D].武汉:中国地质大学, 2012.
[10] 龚芳.反转类指定距离度量的改进及应用研究[D].武汉:中国地质大学, 2021. DOI:10.27492/d.cnki.gzdzu.2021.000301
[11] ZHANG H, PETITJEAN F, BUNTINE W. Hierarchical gradient smoothing for probability estimation trees[C] //Pacific-Asia Conference on Knowledge Discovery and Data Mining, Cham: Springer, 2020: 222-234. DOI: 10.1007/978-3-030-47426-3_18.
[12] QIU C, JIANG L X, LI C Q. Not always simple classification: Learning SuperParent for class probability estimation[J]. Expert Syst Appl, 2015, 42(13): 5433-5440. DOI: 10.1016/j.eswa.2015.02.049.
[13] GONG F, JIANG L X, ZHANG H, et al. Gain ratio weighted inverted specific-class distance measure for nominal attributes[J]. Int J Mach Learn Cyb, 2020, 11(1): 2237-2246. DOI: 10.1007/s13042-020-01112-8.
[14] RENTEA RM, ST PETER SD. Pediatric appendicitis[J]. Surg Clin North Am, 2017, 97(1): 93-112. DOI: 10.1016/j.suc.2016.08.009.
[15] MARCINKEVICˇS R, REIS WOLFERTSTETTER P, KLIMIENE U, et al. Interpretable and intervenable ultrasonography-based machine learning models for pediatric appendicitis[J]. Med Image Anal, 2024, 91: 103042. DOI: 10.1016/j.media.2023.103042.
[16] 吴燕妮.基于人工智能的急性阑尾炎病理类型预测模型建立[D].广州: 南方医科大学, 2022. DOI:10.27003/d.cnki.gojyu.2022.000934. (

基于条件概率分布的混合距离度量方法及应用

Hybrid distance measurement method and application based on conditional probability distribution

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 4

编辑推荐

Metrics

本文评价

[1]	李纯果,张春琴,李海峰. 基于排序互信息的无监督特征选择[J]. 河北大学学报(自然科学版), 2020, 40(2): 200-204.
[2]	毛华, 赵书峰. 最小相关性最大依赖度属性约简的改进算法[J]. 河北大学学报(自然科学版), 2019, 39(3): 225-229.
[3]	李新福,赵杰,梁巍. 基于互信息的宋史语料库词表的提取[J]. 河北大学学报(自然科学版), 2006, 26(5): 557-560.
[4]	王煜,张明,马力. 基于词条聚合和决策树的文本分类方法[J]. 河北大学学报(自然科学版), 2005, 25(3): 338-342.