Abstract: The intrusion detection data contains large redundant and noisy features, as well as some continuous attributes. This paper presents an algorithm based on neighborhood rough set and Particle Swarm Optimization algorithm for the effect of intrusion detection. Training subset is reduced by using neighborhood rough set, and new training subset is generated. The redundant attributes and noise are eliminated to avoid information loss when using traditional rough set; parameters of Support Vector Machine are optimized using particle swarm algorithm to avoid the risk of low precision by subjective choiced parameters. It improves the performance of the intrusion detection. The simulation results in the KDD99 dataset show that the algorithm can effectively improve the accuracy and efficiency of intrusion detection. It has high generalization and stability.

Key words: network intrusion detection, neighborhood rough set, Support Vector Machine（SVM）, Particle Swarm Optimization（PSO） algorithm

摘要： 入侵检测数据往往含有大量的冗余、噪音特征及部分连续型属性，为了提高网络入侵检测的效果，利用邻域粗糙集对入侵检测数据集进行属性约简，消除冗余属性及噪声，也避免了传统粗糙集在连续型属性离散化过程中带来的信息损失；使用粒子群算法优化支持向量机的核函数参数和惩罚参数，以避免靠主观选择参数带来精度较低的风险，进一步提高入侵检测的性能。仿真实验结果表明，该算法能有效提高入侵检测的精度，具有较高的泛化性和稳定性。

关键词: 入侵检测, 邻域粗糙集, 支持向量机, 粒子群算法