Search

Currently, the number of DDoS attacks on devices connected to the Internet has significantly increased. As a part of a study there was implemented an application that classifies network traffic using the neural network approach. Two neural network topologies were developed using SGD and Adam optimizers. A classification accuracy amounted to 99%.

Recently, more attention has been paid to the task of traffic classification. This is due to the fact that, in the era of active development of digital systems, it is necessary to pay particular attention to the issues of application security and the protection of user data). There are many different algorithms and approaches with their own advantages and disadvantages, which differ from each other in processing speed, field of applicability and accuracy of the final results. One of the most actively developing areas at the moment is the use of various machine learning algorithms. This direction also has some disadvantages. In particular, there is a problem in the classification of completely new traffic, which was not during training, the need for periodic retraining when changing the characteristics of network protocols or other important parameters for any particular task. This article describes a network traffic classification model using reinforcement learning methods.

The problem of traffic classification has been studied for a long time: a significant amount of research work has been devoted to its analysis and the search for effective solutions under various conditions and limitations, including in recent years. Classification of network traffic is widely required for many network management tasks, such as stream prioritization, traffic shaping/control, and diagnostic monitoring. When solving network management problems, problems arise such as characterization and modelling of workload, capacity planning, route preparation, and accurate identification of network traffic. This article provides a description of network traffic and an overview of machine learning methods for classifying it, such as Naive Bayes, C4.5, AdaBoost and Random Forest, their advantages, disadvantages and features.

In the modern world of informatization, the volume of information, as well as its transmission over the network, continues to grow. However, network security solutions do not always keep up with industry development. Despite the large number of studies in the field of network security, there are more problems than solutions. Like any other protocol, SNMP is subjected to DDoS attacks, which can cause significant harm to users. The most common attack is the so-called SNMP Amplification. This article presents the architecture of one of the most vulnerable protocols - SNMP, gives a historical overview of the development of this protocol, describes its advantages and disadvantages, and outlines prospective studies in the field of protection of this protocol. For example, several machine learning methods have been described that are used to analyse network traffic at the SNMP level.