Проектирование нейросетевых моделей для обнаружения вторжений с использованием общедоступных баз данных
The article presents a method of constructing a neural network model for detecting and classifying network intrusions using information about attacks contained in the databases of KDD Cup 1999 Data and UNSW-NB-15. Various variants of neural networks with a full and reduced set of input parameters are constructed, a comparative analysis with similar models is performed. A multilayer perceptron MLP 194-20-10 was obtained, using 32 input parameters and capable of distinguishing six types of network connections in the medium with a probability of 98.99%. The error of the first kind was 0.16%; the second kind of error was 4.48%. Based on the obtained model, it is planned to create a software module for intelligent network traffic analysis.