摘要： 在入侵检测应用中，SVM能够在小样本条件下保持良好的检测状态。该文提出了一种改进的SVM方法，其在特定概率指导下删减训练集中的非有效样本，取得了更优的分类效果，改善了传统SVM训练和分类中存在的高资源占用和时耗过高的状况。对DARPA数据的检测实验表明，该方法在入侵检测上有较好的表现。

关键词: 入侵检测, 支持向量机, 缩减训练集

Abstract: In the application of intrusion detection, SVM maintains fine detection status on the condition of small-scale dataset. This paper proposes an improved SVM method. Through cutting non-effective records from training set under the guidance of specific probabilities, it gains better classification results and greatly ameliorates the situation of high resources occupation and time cosumption in traditional SVM training and classification. The tests on DARPA dataset show that this method performs well in intrusion detection.

Key words: intrusion detection, support vector machine(SVM), reduced training set

中图分类号: 

• TP393.08