基于半监督优化分类的入侵检测方法研究

Optimization Method for Semi-supervised Learning in Intrusion Detection

为解决入侵检测训练集(通常包含大量无标记样本和少量已标记样本),在传统半监督支持向量机(S3VM)上确定最优分类决策面,提出一种优化的多分类决策S3VM方法(MLL_S3VM)。该方法结合启发式搜索和聚类方法筛选出差异性较大的分类决策面,采用距离向量法对未标记样本进行标记。实验结果表明,在入侵检测中,该算法明显提高了模型预测精确度。

In order to solve intrusion detection training data set（usually contains a large number of unlabeled samples and a small amount of samples has been marked） on the traditional Semi-Supervised Support Vector Machines（S3VM） to determine the optimal classification decision surface,an optimal S3VM method（MLL_S3VM） based on multiple large-margin low-density separators is proposed.The proposed algorithm combined heuristic sampling search and clustering method,unlabeled examples are estimated using distance vector method at last.Experimental results show that it has better classification accuracy in intrusion detection.