Abstract: This paper proposes a new neural network ensembles method for intrusion detection system. In order to improve the diversity among component networks, different training sets are used for training different networks. On the basis of ensuring the numbers of the component networks, the algorithm selects the networks with proper diversity factor. Experimental results show that the performance of this ensemble method is better than the existing popular. While keeping high detection rate, it has a low false positive rate and is also better for unknown intrusion detection.

Key words: network security, intrusion detection, anomaly detection, artificial examples training, neural network ensembles, diversity factor

摘要： 提出一种基于人工示例训练的神经网络集成入侵检测方法。使用不同的训练数据集训练不同的成员网络，以此提高成员网络之间的差异度。在保证成员网络个数的基础上，选择差异度较大的成员网络构成集成，以提高系统的整体性能。实验结果表明，与当前流行的集成算法相比，该方法在保证较高入侵检测率的前提下，可保持较低的误检率，并对未知入侵也具有较高的检测率。

关键词: 网络安全, 入侵检测, 异常检测, 人工示例训练, 神经网络集成, 差异度

CLC Number: 

• TP393