基于邻域粒化的逻辑回归算法  

Logistic Regression Algorithm Based on Neighborhood Granulation

作　　者：闫静茹 陈颖悦[1] 曾高发 刘培谦 傅兴宇 YAN Jingru;CHEN Yingyue;ZENG Gaofa;LIU Peiqian;FU Xingyu(School of Economics and Management,Xiamen University of Technology,Xiamen 361024,China;Xiamen Zhixiang Intelligent Technology Co.,Ltd.,Xiamen 361000,China;College of Computer and Information Engineering,Xiamen University of Technology,Xiamen 361024,China)

机构地区：[1]厦门理工学院经济与管理学院,福建厦门361024 [2]厦门市执象智能科技有限公司,福建厦门361000 [3]厦门理工学院计算机与信息工程学院,福建厦门361024

出　　处：《山西大学学报（自然科学版）》2024年第1期40-47,共8页Journal of Shanxi University(Natural Science Edition)

基　　金：厦门市科技计划项目(2022CXY0428)。

摘　　要：逻辑回归作为一种经典的分类算法,其结构简单且可解释性强。然而,逻辑回归难以处理模糊与不确定的非线性数据。为了解决这一问题,通过采用粒计算理论中的邻域粒化技术,提出了一种基于邻域粒化的逻辑回归算法。对于非线性数据,邻域粒化使数据更容易进行分离和构造。首先,对数据集样本的单特征进行邻域粒化,构造出邻域粒子。然后在多特征上形成邻域粒向量。此外,定义了这些邻域粒向量的度量与运算规则,并设计了一种邻域粒逻辑回归算法,有效地提高了逻辑回归的分类准确性。在WDBC(Diagnostic Wisconsin Breast Cancer),Iris以及Seeds等数据集上进行了分类实验,与经典的逻辑回归进行了比较,结果表明,本文提出算法的分类准确率相较于经典的逻辑回归在三个数据集上分别高出0.6%,7.6%,4.1%。As a classical classification algorithm,logistic regression has a simple structure and strong interpretability.However,logistic regression is difficult to deal with fuzzy and uncertain nonlinear data.To solve this problem,a logistic regression algorithm based on neighborhood granulation is proposed by using neighborhood granulation technology in granular computing theory.For nonlinear data,neighborhood granulation makes the data easier to separate and construct.Firstly,the neighborhood granules are constructed by the neighborhood granulation of the single feature of the data set sample.The neighborhood granular vectors are then formed on the multi-feature.In addition,the measurement and operation rules of these neighborhood granular vectors are defined,and a neighborhood granular logistic regression algorithm is designed,which effectively improves the classification accuracy of logistic regression.Classification experiments are carried out on WDBC(Diagnostic Wisconsin Breast Cancer),Iris and Seeds data sets,and compared with the classical logistic regression.The results show that the classification accuracy of the proposed algorithm is 0.6%,7.6%and 4.1%higher than that of the classical logistic regression in the three data sets,respectively.

关 键 词：逻辑回归 单特征粒化 粒计算 邻域粒子 粒向量 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]