Machine Learning Aspects of Internet Firewall Data

One of the numerous implementations of neural networks (NN), as part of machine learning, is the modeling of network firewall rules. For this purpose, a suitable dataset containing standard traffic attributes as well as the ability of Weka software package for modeling and testing multilayer perceptrons (MLP) was used. The aim of this paper was to create and examine an NN model of Internet firewall and optimize its parameters that best simulates the operation of rules. It was found that the number of neurons in hidden layers, the learning rate, momentum, and number of epochs affect the accuracy, while the impact of percentage split and batch size can be ignored. Also, it performed an evaluation of losses of different activation functions in NN environment, with previously determined optimal parameters. Moreover, it has been shown that the following algorithms provided the highest accuracy in solving classification problems for a firewall dataset: Random Forest, J48 and MLP. From the aspect of the possibility of clustering firewall data, the paper found that the k-means algorithm showed greater accuracy and speed than the EM and DBSCAN algorithms.