采用自适应GHA神经网络的分类器设计

介绍了一种自适应逼近数据实质维的GHA神经网络学习算法。基于主元子空间分解的思想,给出了基于该算法的分类器刻画方法,对其中的刻画参数给出了详细的界定。该分类器采用监督学习机制进行训练,可以自动学习输入的主元特征子空间维数。在入侵检测领域,利用KDD CUP 1999数据集对该方法进行了仿真。采用正常连接数据训练GHA异常检测分类器,利用拒绝服务攻击数据进行了误用检测训练。并将测试结果与其他入侵检测方法进行了比较。

Classifier Design Using Adaptive GHA Neural Networks

An adaptive Generalized Hebbian Algorithm (GHA) is presented which can be used to approach the intrinsic dimension of an input data set. The classifier design based on adaptive GHA networks is given in detail and the determination method of the classifier parameters is also described. The classifier can be trained by using supervised manner. We applied this approach to the domain of intrusion detection. Some simulations are carried out for anomaly detection by using labeled normal type network connections, and the misuse detection are performed on specified type attacks of denial-of-service intrusions. All the training and testing datasets are based on the KDD CUP 1999 intrusion evaluation data set. Performance comparisons are also made with other recent published methods.