微型电脑应用

采用集成学习模式进行入侵检测时,可以获得比单个学习器更高效的网络攻击识别过程,并能显著提高识别准确率。设计的SN通过MDM对各KELM子学习器计算得到集成增益度,从中选出具有较高增益度的KELM子学习器再实施集成。选择Bagging方式完成抽样集成过程,同时以Hadoop分布式结构对算法实施训练,通过并发方式完成各子KELM的检测,使算法达到更高的效率。通过测试发现,不管对于公共KDD99数据集还是以手工方式建立的网络物理仿真系统,SN都可以高效发现各类入侵行为,满足实际应用要求。

The integrated learning mode can obtain a more efficient network attack identification process than a single learner, and it significantly improves the recognition accuracy. The SN designed in this paper calculates the integration gain of each KELM sub-learning device through MDM, and then selects the KELM sub-learning device with higher gain degree to implement integration. The bagging is selected to complete the sampling integration process. Meanwhile, the algorithm is trained by Hadoop distributed structure, and the detection of each sub-KELM is completed by means of concurrency, so as to achieve higher efficiency of the algorithm. Through testing, it is found that the SN can efficiently detect various intrusion behaviors for both public KDD99 data set and manual network physical simulation system, which meets the practical application requirements.