Multilayer Perceptron: an Intelligent Model for Classification and Intrusion Detection

Nowadays the security of computer devices is growing significantly. This is due to more and more devices are connected to the network. For this reason, optimize the performance of systems able to detect intrusions (IDS) is a goal of common interest. The following work consists of use the generalizing power of neural networks to classify the attacks. In particular, we will use multilayer perceptron (MLP) with the algorithm of back-propagation algorithm and the sigmoidal activation function. We use a subset of the DARPA dataset, known as KDD99. It is a public dataset labeled for an IDS and previously processed. We will make an analysis of the results obtained using different configurations, varying the number of hidden layers and the number of training epochs to obtain a low number of false results. We observe that it is required a large number of training epochs and how, using the entire data set consists of 41 features, the best classification is carried out for the type of DOS and Probe attacks.