论文部分内容阅读
提出了一种构建轻量级的IP流分类器的wrapper型特征选择算法MRMHCLSVM。该算法采用改进的随机变异爬山(MRMHC)搜索策略对特征子集空间进行随机搜索,然后利用提供的数据在无约束优化线性支持向量机(LSVM)上的分类错误率作为特征子集的评价标准来获取最优特征子集。在IP流数据集上进行了大量的实验,实验结果表明基于MRMHC-LSVM的流分类器在不影响分类准确度的情况下能够提高检测速度,与当前典型的流分类器NBK-FCBF相比,基于MRMHC-LSVM的IP流分类器具有更小的计算复杂度与更高的检测率。
A wrapper-based feature selection algorithm MRMHCLSVM is proposed to construct a lightweight IP flow classifier. This algorithm uses a modified random mutation climbing (MRMHC) search strategy to search the feature subset space randomly, and then uses the classification error rate of the provided data on the unconstrained optimized linear support vector machine (LSVM) as the evaluation criterion of the feature subset To get the best subset of features. A large number of experiments on IP stream data sets show that the MRMHC-LSVM-based stream classifier can improve the detection speed without affecting the classification accuracy. Compared with the current typical classifier NBK-FCBF, The IP flow classifier based on MRMHC-LSVM has less computational complexity and higher detection rate.