To search, Click below search items.


All Published Papers Search Service


Feature Deduction and Ensemble Design of Parallel Neural Networks for Intrusion Detection System


Syed Muhammad Aqil Burney, M. Sadiq Ali Khan


Vol. 10  No. 8  pp. 259-267


In this modern age of computer networks, there is an ultimate demand for development of reliable, extensible, easily manageable and have low maintenance cost solutions for intrusion detection. We have used KDD’99 dataset for experimental verifications of our proposed approach. With the features reduction step, it is possible to significantly reduce the number of input features so that the chance of over-fitting and data redundancy can be reduced. Then a multilayer Perceptron neural network classifier is applied on the selected feature space using one-against-one approach. For the training of neural network, each attack is trained with the normal dataset. Thus we have four neural networks working in parallel such as normal vs. probe, normal vs. DoS, normal vs. U2R and normal vs. R2L. After repeated simulations and bootstrapping, we have shown that our proposed approach has good results for Probe, DoS and R2L attacks and average results for U2R attack. A comparison with other intrusion detection systems is also presented.


Intrusion Detection System (IDS), KDD-cup dataset, principal component analysis (PCA), neural networks (NN), ensemble of one-against-one approach