基于动态分类算法的入侵检测系统

Intrusion Detection System Based on Dynamic Classification Algorithm

为了使分类方法适合网络入侵检测系统在线、实时的特点,根据自适应谐振理论提出了基于联想和竞争学习的动态分类算法。算法采用改进的胜者全得方法训练神经网络,通过估算类别数目方法优化初始系数。该算法能防止分类时丢弃特殊模式,并能对未知类别数量的数据动态进行分类,实验采用相同的数据集训练自适应谐振理论网络和动态分类网络。结果证明,动态分类算法具有良好的收敛性和模式分类能力。将该算法应用于入侵检测系统的网络行为智能学习,其误报率仅为10%左右。

A network intrusion detection system based on dynamic classification is described and a new competitive algorithm is given. Through analyzing adaptive resonance theory, a dynamic classification algorithm based on associative and competitive learning is provided. The algorithm uses an improved winner taking all method to train the network. Before training an estimating method is used to initialize parameters more correctly. The algorithm can prevent from discarding irregular data when the number of group is fixed and can classify the patterns when the number of group is unknown. In the experiments, the same data are used to train the adaptive resonance theory net and the dynamic classification algorithm net. The results of experiments show the capability and effectiveness of the dynamic classification algorithm. When applying the algorithm to network behavior intelligent learning of intrusion detection system, the error rate is about 10%.