基于粒计算的逻辑回归分类方法  

Logistic Regression Classification Method Based on Granular Computing

作　　者：廖亮 陈颖悦 曾高发 刘培谦 LIAO Liang;CHEN Ying-yue;ZENG Gao-fa;LIU Pei-qian(College of Computer and Information Engineering,Xiamen University of Technology,Xiamen Fujian 361024,China;Xiamen Zhixiang Intelligent Technology Co.,Ltd,Xiamen Fujian 361024,China;Economy and management,Xiamen University of Technology,Xiamen Fujian 361024,China)

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

出　　处：《计算机仿真》2025年第2期382-388,共7页Computer Simulation

基　　金：国家自然科学基金项目(61976183);2022年厦门市科技计划项目(2022CXY0428)。

摘　　要：逻辑回归算法属于监督学习,具有实现简单、速度快的特点,是一种广泛使用于工业应用的分类算法。然而,在处理非线性可分与特征空间大数据量小的数据集时,其分类效果不佳。针对非线性准确率不高和特征空间大导致欠拟合的问题,提出一种基于粒计算的逻辑回归分类方法。通过引入粒计算理论,训练样本在单特征上使用粒化技术构造粒子,在多特征上使用粒化技术形成粒向量,决策特征使用异或粒化技术形成决策粒子。最后,在三个Kaggle数据集和两个UCI数据集上进行实验。将上述方法与逻辑回归、朴素贝叶斯、决策树三种经典分类算法以及XGboost、LightGBM两种最新分类算法进行了比较。其结果表明基于粒计算的逻辑回归分类方法的可行性和有效性。Logistic regression algorithm belongs to supervised learning,which has the characteristics of simple implementation and fast speed.It is a kind of classification algorithm widely used in industrial applications.However,the classification effect is not good when dealing with nonlinear separable data sets with large feature space and small amount of data.A logistic regression classification method based on granular computing was proposed to address the problem of underfitting caused by low nonlinear accuracy and large feature space..By introducing granular computing theory,granulation technology was used to construct particles on single feature for training samples,granulation technology was used to form granular vectors on multiple features,and XOR granulation technology was used to form decision particles on decision features.Finally,some experiments were conducted on three Kaggle data sets and two UCI data sets.The proposed method was compared with three classical classification algorithms,namely logistic regression,naive Bayes and decision tree,and two new classification algorithms,namely XGboost and LightGBM.The results show that the logistic regression classification method based on granular computing is feasible and effective.

关 键 词：粒计算 异或 逻辑回归 监督学习 粒向量 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]