网络入侵检测的机器学习算法评估与比较

摘要

Traditional intrusion detection method is unable to meet the needs of big-data era in accuracy and effectiveness, while the machine-learning algorithm is becoming main-stream. Now the main research focuses on machine-learning algorithm in support vector machines, and however it also has its own defects. Therefore, other excellent classification algorithms in machine learning are introduced. In addition, the classical NSL-KDD data set is used to compare the accuracy of the algorithm, and the applicable environment analyzed, thus to provide a basis for intrusion detection analysis in different scenarios in the future. After using the data set to complete the model training, the ROC curve, accuracy and other indicators are used to evaluate the model, and fairly good results are acquired.%入侵检测传统方法的准确性和有效性已经无法满足大数据时代的需求,机器学习算法日趋成为主流.现主要研究侧重于机器学习算法中的支持向量机,但其也有自身的缺点.因此,引入其他机器学习中的其他优秀分类算法,并使用经典的NSL-KDD数据集对比算法的准确性,分析适用环境,为将来不同场景下的入侵检测分析提供基础.在使用数据集完成模型训练后,使用ROC曲线、准确率等指标对模型进行评估,得出了较好的结果.