«上一篇/Previous Article|本期目录/Table of Contents|下一篇/Next Article»

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

34

期数:

2024年07期

页码:

100-107

栏目:

网络空间安全

出版日期:

2024-07-10

文章信息/Info

Title:

An Intrusion Detection Approach Incorporating SMOTE-Tomek Link with Integrated Modeling

文章编号:

1673-629X(2024)07-0100-08

作者:

李润杰; 张小庆*; 刘昌华

武汉轻工大学 数学与计算机学院,湖北 武汉 430048

Author(s):

LI Run-jie; ZHANG Xiao-qing*; LIU Chang-hua

School of Mathematics and Computer Science,Wuhan Polytechnic University,Wuhan 430048,China

关键词:

入侵检测; 采样算法; 特征选择; 集成学习; 网络安全

Keywords:

intrusion detection; sampling algorithms; feature selection; ensemble learning; cyber security

分类号:

TP309

DOI:

10.20165/j.cnki.ISSN1673-629X.2024.0105

摘要:

随着全球互联网的快速扩张和网络安全威胁日益复杂化,开发高效稳定的入侵检测系统成为了网络安全领域中的重要研究任务。 该文的重点在于解决入侵检测数据集普遍存在的两个难题:一是由于正常与异常网络行为样本数量差异引起的类别不均衡;二是由于数据集中冗余和无效特征过多导致的高维度问题。 为此,运用集成学习思想,融合 SOMTE-Tomek Link 综合采样算法与 3 个同质模型,提出针对非平衡数据集的加权投票法集成模型。 先利用 SOMTE-Tomek Link 综合采样算法对非平衡数据进行预处理,再利用随机森林排列重要性度量算法对数据集进行有效特征选择,以降低模型检测失误率与计算开销。 将多个机器学习模型和所提集成模型进行对比实验评估。 实验结果表明,集成模型准确率达到 97. 84% ,比单一模型可提高 1 ~ 4 百分点,在少样本攻击分类的准确率、精确率、召回率和 F1 分数上有大幅提升,且模型训练效率更高,稳定性更强。

Abstract:

With the rapid expansion of the global Internet and the increasing complexity of network security threats,the development of ef-ficient and stable intrusion detection systems has become an important research task in the field of network security. The focus of this paper is to solve two common problems in intrusion detection data sets:one is the category imbalance caused by the difference in the number of normal and abnormal network behavior samples;The second is the high dimensional problem caused by too many redundant and invalid features in the data set. Therefore,according to ensemble learning,by combining the SOMTE-Tomek Link comprehensive sampling algorithm with three homogeneous models,an integrated weighted voting model for unbalanced data sets is proposed. The SOMTE-Tomek Link algorithm preprocesses the data,while the random forest ranking importance measure algorithm selects effective features,reducing error rates and computational demands. Comparative evaluations against multiple machine learning models demonstrate that the integrated model achieves a 97. 84% accuracy rate, outperforming single models by 1 ~ 4 percentage points. Notably, it significantly improves accuracy, precision, recall, and F1 scores in classifying less - sample attacks, enhancing training efficiency and model stability.

常用功能

更新日期/Last Update: 2024-07-10