点击复制
一种混合特征选择的朴素贝叶斯网络入侵检测算法(
)
《南京师大学报（自然科学版）》[ISSN:1001-4616/CN:32-1239/N]
- 卷:
-
48
- 期数:
-
2025年03期
- 页码:
-
73-83
- 栏目:
-
计算机科学与技术
- 出版日期:
-
2025-06-20
文章信息/Info
- Title:
-
A Naive Bayes Network Intrusion Detection Algorithm with Mixed Feature Selection
- 文章编号:
-
1001-4616(2025)03-0073-11
- 作者:
-
郑锦波; 王慧玲
-
(伊犁师范大学网络安全与信息技术学院,新疆 伊宁 835000)
- Author(s):
-
Zheng Jinbo; Wang Huiling
-
(School of Network Security and Information Technology,Yili Normal University,Yining 835000,China)
-
- 关键词:
-
- Keywords:
-
- 分类号:
-
TP309
- DOI:
-
10.3969/j.issn.1001-4616.2025.03.009
- 文献标志码:
-
A
- 摘要:
-
在入侵检测应用中,机器学习算法发挥着至关重要的作用,特征选择作为关键的数据预处理步骤,可以有效提升分类器的分类效果. 而现有的特征选择算法未考虑数据分布不均匀时特征间存在的伪相关性,影响了分类器的泛化能力. 针对此问题,本文提出了一种混合特征选择的朴素贝叶斯网络入侵检测算法,将相关性度量准则引入特征提取阶段,避免特征间存在的伪相关性,更好地满足朴素贝叶斯算法的强假设,使模型检测性能有效提升. 该方法采用了两步特征选择策略:第一步筛选数据集中和类变量相关性较强特征; 第二步去除冗余特征,筛选出相互条件独立的特征作为特征子集,并将此特征子集送入朴素贝叶斯算法进行检测. 实验结果表明,提议的方法在检测率和泛化性能上都优于参与对比的6个传统机器学习算法,并且在一定程度上克服了数据分布不平衡导致的精度低的问题,与近期提出的两个深度学习算法相比较,在准确率和精确率上优于两个对比深度学习算法.
- Abstract:
-
In intrusion detection applications,machine learning algorithms play a crucial role. Feature selection,as a key data preprocessing step,can effectively improve the classification performance of classifiers. However,existing feature selection algorithms do not consider the existence of pseudo-correlations between features when the data distribution is imbalanced,which affects the generalization ability of classifiers. To address this issue,a hybrid feature selection naive Bayes network intrusion detection algorithm is proposed,which introduces correlation measurement criteria into the feature extraction stage to avoid the pseudo-correlations between features and better satisfy the strong assumption of the naive Bayes algorithm,thereby improving the detection performance of the model. This method adopts a two-step feature selection strategy. In the first step,features that are strongly correlated with the class variable are selected from the dataset. In the second step,redundant features are removed to select a subset of mutually conditionally independent features,which are then fed into the naive Bayes algorithm for detection. Experimental results show that the proposed method outperforms 6 traditional machine learning algorithms in terms of detection rate and generalization performance,and it partially overcomes the problem of low accuracy caused by imbalanced data distribution. Compared with two recently proposed deep learning algorithms,it performs better in terms of accuracy and precision.
更新日期/Last Update:
2025-06-20