摘要
随着网络攻击手段日趋多样化和复杂化,传统的网络安全防护体系例如以防火墙为代表的边界防护技术以及基于误用检测的统计学习方法等正面临严峻挑战,在应对新型网络安全威胁时已显现出明显的局限性,其防护效果和响应能力均难以满足当前网络安全防护需求。论文提出一种基于受限玻尔兹曼机(RBM)和三支决策(3WD)的入侵检测方法。选取RBM对数据集进行特征提取,在分类决策阶段,针对传统二分类决策中“非此即彼”的局限性,引入三支决策理论对决策机制进行优化。在保持原有正域(1)和负域(0)划分的基础上,通过设定合理的阈值区间,将处于决策模糊带的样本归入边界域。不仅能够有效处理不确定性样本,通过边界域的设置降低决策风险,提高分类系统的容错性和实用性。实验评估结果显示,在广泛使用的NSL-KDD基准数据集上该方法的准确率为96.1%,高于对比方法中准确率最高的基于时空特征的分层入侵检测系统2.5%。
With the increasing diversification and sophistication of cyberattack techniques,traditional cybersecurity defense systems,such as perimeter protection technologies exemplified by firewalls and statistical learning methods based on misuse detection,are facing significant challenges.These systems have exhibited notable limitations in addressing emerging cybersecurity threats,as their protective efficacy and response capabilities fall short of meeting current cybersecurity requirements.This paper proposes an intrusion detection method based on restricted Boltzmann machines(RBM)and three-way decisions(3WD).Specifically,RBMis utilized for feature extraction from the dataset,while the three-way decision theory is introduced during the classification decision phase to optimize the decision-making mechanism.This approach addresses the inherent"either-or"limitation of traditional binary classification by incorporating a boundary domain.By retaining the original positive(1)and negative(0)domain divisions and establishing appropriate threshold intervals,samples within the decision ambiguity zone are assigned to the boundary domain.This not only enhances the handling of uncertain samples but also reduces decision risks and improves the fault tolerance and practicality of the classification system.Experimental evaluations conducted on the widely used NSL-KDD benchmark dataset demonstrate that the proposed method achieves an accuracy of 96.1%,surpassing the highest accuracy among comparative methods by 2.5%,specifically outperforming a spatiotemporal feature-based hierarchical intrusion detection system.
作者
洪亮
HONG Liang(College of Computer,Jiangsu University of Science and Technology,Zhenjiang 212100)
出处
《计算机与数字工程》
2025年第6期1686-1691,1739,共7页
Computer & Digital Engineering
关键词
入侵检测
玻尔兹曼机
三支决策
特征提取
网络安全
intrusion detection
Boltzmann machines
three-way decision
feature extraction
network security
