模糊支持向量机在入侵检测中的应用研究

摘要

网络入侵检测是保证安全防护技术,在入侵检测中,数据分布的不均衡和噪声数据的存在影响检测性能和分类效果.针对传统支持向量机对噪声数据和孤立点敏感的缺点,提出了一种基于双超球隶属度函数的模糊支持向量机算法.算法在确定隶属度时充分考虑样本与类中心之间的关系以及类中各个样本之间的关系,并且将样本的隶属度与样本到所在类中心的距离看作是一个非线性关系.根据模糊支持向量机和双超球隶属度函数的原理,采用核函数对检测性能的影响.通过KDD99数据的测试并与传统的支持向量机算法进行比较,实验结果证明改进算法的可行性和有效性.%The classification results affected by uneven distribution of data in intrusion detection, a new Fuzzy Support Vector Machine method based on double hypersphere membership function is proposed against the sensitivity of traditional Support Vector Machine to noisy data and outlier. Relationship between samples and centers and relationship among samples are considered when membership is computed. The relationship between the membership of the sample and the distance from its class center is considered to be nonlinear. The principles of Fuzzy Support Vector Machine and double hypersphere membership function are illustrated in the article. The choice of parameter of Kernel function is given to illustrate the performance. The results show the feasibility and effectiveness through tests with DARPA KDD99 and comparison between the algorithms.