Intrusion detection and classification is an important process in securing information in the network supported computers. Due to the popular access of internet resources, by using enormous access methods, malicious activities are done through internet to add information, to delete information and to prevent working of the computers. In spite of various attacks prevention soft wares, new powerful attacks are generated. An approach has been developed to overcome such new unknown attacks. This paper has attempted to implement Fisher's linear discriminant function (FLD) and artificial neural network (ANN) with supervised radial basis function has been implemented. A set of knowledge discovery and data mining (KDD) cup network intrusion dataset of the year 1999 have been obtained. This data set has been separated using variance analysis into training (183 patterns) and testing (2973 patterns) The FLD algorithm takes the inputs of the training data (183 X 41) and outputs φ1(41 X 1) and φ2(41 X 1) discriminant vectors. The training data is further inner producted with φ1 and φ2 to obtain training patterns (183 X 2) for training the Radial basis function neural network. A set.of final weights are obtained from the RBF (3 X 1) when the number of cluster centre is considered as 2 in the hidden layer of the RBF. The topology of the RBF is {2 (input layer) X 2 (hidden layer) X 1(output layer)}. The attack types considered are snmpgetattack and smurf.
Fisher's linear discriminant function, Radial basis function, ndimensional transformation
