摘 要
随着信息技术的迅猛发展，网络安全威胁日益复杂多变，传统入侵检测系统在应对新型攻击时面临诸多挑战。为此，本研究旨在构建基于机器学习的网络安全入侵检测系统，以提高检测效率和准确性。研究选取了多种典型机器学习算法，包括支持向量机、随机森林、深度神经网络等，并对KDD CUP 99、NSL - KDD等公开数据集进行预处理，通过特征选择与提取优化输入特征空间。实验结果表明，相较于传统方法，所提系统在检测率、误报率等关键性能指标上均有显著提升，其中深度神经网络模型表现尤为突出，在处理大规模、高维度数据时展现出强大的泛化能力。该研究创新性地将迁移学习引入入侵检测领域，有效解决了不同网络环境下模型适应性问题，为跨平台应用提供了理论依据和技术支持。此外，还提出了一种基于主动学习的样本标注机制，降低了人工标注成本的同时提高了模型训练效率，对于推动机器学习技术在网络安全领域的深入应用具有重要意义。
关键词：机器学习；入侵检测系统；深度神经网络；迁移学习；主动学习
Abstract
With the rapid development of information technology, cyber security threats have become increasingly complex and diverse, posing significant challenges to traditional intrusion detection systems in. To address these challenges, this study aims to construct a machine learning-based network security intrusion detection system to improve detection efficiency and accuracy. Multiple representative machine learning algorithms, including Support Vector Machines (SVM), Random Forests, and Deep Neural Networks (DNN), were selected for this research. Public datasets such as KDD CUP 99 and NSL-KDD were preprocessed, and feature selection and extraction were employed to optimize the input feature space. Experimental results demonstrate that, compared with traditional methods, the proposed system achieves significant improvements in key performance metrics such as detection rate and false positive rate, with the DNN model showing particularly outstanding performance in handling large-scale, high-dimensional data and exhibiting strong generalization capabilities. This study innovatively introduces transfer learning into the field of intrusion detection, effectively solving the model adaptability issue across different network environments and providing theoretical basis and technical support for cross-platform applications. Additionally, an active learning-based sample labeling mechanism is proposed, which reduces the cost of manual labeling while improving model training efficiency, contributing significantly to the in-depth application of machine learning technologies in the domain of cyber security.
Keywords：Machine Learning；Intrusion Detection System；Deep Neural Network；Transfer Learning；Active Learning
目 录
摘 要 I
Abstract II
引 言 1
第一章 网络安全入侵检测系统概述 2
1.1 入侵检测系统发展历程 2
1.2 传统入侵检测方法局限性 2
1.3 机器学习在入侵检测中的优势 3
第二章 机器学习算法选择与优化 4
2.1 常用机器学习算法比较 4
2.2 特征选择与数据预处理 4
2.3 模型训练与性能评估 5
第三章 入侵检测系统的架构设计 7
3.1 系统整体架构规划 7
3.2 数据采集与预处理模块 7
3.3 异常检测与响应机制 8
第四章 实验验证与结果分析 10
4.1 实验环境与数据集构建 10
4.2 不同算法对比实验 10
4.3 系统性能评估与改进 11
结 论 13
参考文献 14
致 谢 15
