Inner parameters' optimization in the artificial neural network for the traffic data classification in radiofrequency applications: Classification of nonstationary data using the machine learning algorithm “random forest”

摘要

The number of devices connected to the World Wide Web and the requirements of subscribers to the speed of mobile Internet access are increasing every year. Developers of telecom equipment and telecom operators, trying to meet new challenges, are preparing to seriously change the architecture of networks and interaction regulations in radiofrequency applications. Availability of required radiofrequencies in radiofrequency applications is one of the main necessary factors for the development of such networks, along with the readiness of the network architecture and infrastructure, business models and subscriber devices [1]. In the development of fifth-generation technologies in radiofrequency applications, there is a certain set of “input” characteristics that serve as a guide for a new standard. For example, compared to the best existing LTE networks, the data transfer speed in 5G networks should be 10-100 times higher, the response time is 5 times less, the network should support the number of devices 100 times more[2-4]. The modern telecommunication market is at a stage when operators have a favorable opportunity to bypass all the convergence difficulties inherent in the networks of the past, and go directly to the next-generation networks based on technology, which received the working name NGN - “New Generation Network”. In order to make this breakthrough and join the number of high-tech operators, new solutions are needed to be constructed in the field of creating and providing high-performance services. NGN - technology of building a network - is designed to provide data transmission services and voice services in radiofrequency applications. It removes a number of restrictions and barriers that exist now, and this is its economic productivity[4]. DPI (Deep Packet Inspection) - technology for the classification and filtering of traffic by its content. DPI is able to define application protocols (HTTP, HTTPS, Skype, BitTorrent), the type of data transferred (web pages, audio and video files), it can extract information specific to specific protocols (URL for HTTP, domain name for HTTPS, distribution identifier BitTorrent). At the moment, most Russian providers use DPI as a means of blocking websites from a single register of prohibited information. The use of DPI for this purpose allows providers to block specific links made to the registry, rather than an IP address or a whole domain[1-4]. Having enough data about the traffic data related to the user, it is possible to successfully determine the thematic categories reflecting the interests and preferences of this user or detect risky activities. In the process of choosing the optimal model for traffic data analisys the so-called “random forest” algorithm was chosen due to the accuracy quality and equation stability: a large number of parameters affecting the implementation of the task are taken into account for solving the classification problem[2].