面向入侵检测的基于IMGA和MKSVM的特征选择算法

入侵检测系统处理的数据具有数据量大、特征维数高等特点,会降低检测算法的处理速度和检测效率。为了提高入侵检测系统的检测速度和准确率,将特征选择应用到入侵检测系统中。首先提出一种基于免疫记忆和遗传算法的高效特征子集生成策略,然后研究基于支持向量机的特征子集评估方法。并针对可能出现的数据集不平衡造成的特征子集评估能力下降,以黎曼几何为依据,利用保角变换对核函数进行修改,以提高支持向量机的分类泛化能力。实验仿真表明,提出的特征选择算法不仅可以提高特征选择的效果,而且在不平衡数据集上具有更好的特征选择能力。还表明,基于该方法构建的入侵检测系统与没有运用特征选择的入侵检测系统相比具有更好的性能。

Feature Selection Algorithm Based on IMGA and MKSVM to Intrusion Detection

In order to improve performances of intrusion detection system in terms of detection speed and detection rate,itis necessary to apply feature selection in intrusion detection system. Firstly,an efficient search procedure based on immune memory and genetic algorithm (IMGA) was proposed. Then, support vector machine (SVM) based on wrapper feature evaluation methods was surveyed,in order to improve the feature selection performance of unbalanced datasets. We used the conformal transformation and Riemannian metric to modify kernel function, and reconstructed a new Modified Kernel SVM (MKSVM). Finally, the simulation experimental results show that this approach can improve the process of selecting important features, and has better feature selection ability on the unbalanced data. Furthermore, the experiments indicate that intrusion detection system with this feature selection algorithm has better performanccs than that without feature selection algorithm.