西安电子科技大学学报 ›› 2023, Vol. 50 ›› Issue (4): 132-138.doi: 10.19665/j.issn1001-2400.2023.04.013

• 网络空间安全专栏 • 上一篇    下一篇

人工鱼群特征选择的网络入侵检测系统

刘景美(),闫义博()   

  1. 西安电子科技大学 通信工程学院,陕西 西安 710071
  • 收稿日期:2023-01-15 出版日期:2023-08-20 发布日期:2023-10-17
  • 通讯作者: 闫义博
  • 作者简介:刘景美(1979—),女,副教授,E-mail:jmliu@mail.xidian.edu.cn
  • 基金资助:
    陕西省自然基金(2021JM-134)

Artificial fish feature selection network intrusion detection system

LIU Jingmei(),YAN Yibo()   

  1. College of Communications Engineering,Xidian University,Xi’an 710071,China
  • Received:2023-01-15 Online:2023-08-20 Published:2023-10-17
  • Contact: Yibo YAN

摘要:

入侵检测领域中,数据的冗余和无关特征不仅减缓了分类的过程,而且会妨碍分类器做出准确的决策,导致入侵检测系统性能下降。针对入侵检测高维数据集带来的系统准确率较低的问题,提出人工鱼群特征选择的网络入侵检测系统。首先对原始数据集预处理,对数据进行清洗并标准化;然后结合自适应参数变化和多目标优化算法,提出一种改进的多目标人工鱼群算法,通过动态优化搜索空间,提升搜索能力,选择最优的特征子集;最后提出一种基于遗传算法和CatBoost的改进多目标人工鱼群优化方法的入侵检测模型,对生成的多组特征子集输入CatBoost进行分类并进行特征评估,检验特征选择的有效性。通过在NSL-KDD数据集上验证,提出的特征选择算法使用17维特征得到约93.97%的准确率,在UNSW-NB15数据集上,算法使用24维特征得到约95.06%的准确率。仿真结果表明,所提算法在维度低的同时可获得高准确率,与现有特征选择方法相比具有一定优势。

关键词: 入侵检测, 特征选择, 人工鱼群, 多目标优化

Abstract:

In the field of intrusion detection,redundancy and extraneous features not only slow down the classification process,but also prevent the classifier from making accurate decisions,resulting in intrusion detection system performance degradation.A network intrusion detection system based on artificial fish feature selection is proposed to address the problem of low system accuracy induced by high-dimensional data sets in intrusion detection.First,the original data set is preprocessed,with the data cleaned and standardized.Then,an improved multi-objective artificial fish swarm algorithm(AFSA) is presented by merging the adaptive parameter modifications and the multi-objective optimization algorithm.By dynamically optimizing the search space,the search ability is improved,and the optimal feature subset is selected.Finally,an intrusion detection model is established based on a genetic algorithm and CatBoost improved multi-objective artificial fish swarm optimization approach.The generated multi-feature subsets are classified by CatBoost for feature evaluation,and the effectiveness of feature selection is tested.The proposed feature selection approach employs 17-dimensional features to achieve an accuracy of 93.97% on the NSL-KDD dataset,while it uses 24-dimensional features to achieve an accuracy of 95.06% on the UNSW-NB15 dataset.Simulation results show that the proposed algorithm can achieve a high accuracy while having a low dimension,which has certain advantages compared with existing feature selection methods.

Key words: intrusion detection system, feature selection, artificial fish swarm algorithm, multi-objective optimization

中图分类号: 

  • TP393