HG AR-PNN在网络入侵检测中的应用

Application of HG AR-PNN Classifier in Network Intrusion Detection

人工神经网络(ANN)大大提高了入侵检测系统的检测性能,但对于出现次数较少的攻击,ANN并不能提供令人满意的稳定性和检测率。提出了一种基于超图Helly性质和算术取余概率神经网络(HG AR-PNN)的入侵检测新方法。该方法利用超图的Helly性质选取最优特征子集,再对最优特征子集进行归一化算术取余,然后实现概率神经网络对数据集的训练。最后,使用KDDCUP’99数据集进行实验,并对HG AR-PNN算法的性能进行评价。实验结果表明,对于不常出现的攻击,HG AR-PNN分类器同样有着较好的稳定性和较高的分类精度。

The learning model which is based on artificial neural network（ANN） can greatly improve the performance of intrusion detection system, but to the less frequent attacks, the ANN can not provide stability and satisfactory detection rate. A new intrusion detection method based on hypergraph Helly property and arithmetic residue probability neural network（HG AR-PNN） is proposed. This method uses hypergraph Helly property to select the best feature subset, and then normalize the optimal feature subset with arithmetic residue, after that, uses the PNN for training the data set. Finally, experiments are carried out using KDDCUP ＇99 data set, and the performance of HG AR-PNN algorithm is evaluated. The experimental results show that HG AR-PNN classifier has better stability and higher classification accuracy for less frequent attacks.