IoT traffic prediction with Neural networks learning based on SDN infrastructure

Artem Volkov, Ali R. Abdellah, Ammar Muthanna, Andrey Koucheryavy
In recent time, there are more achievements in technologies for 5G/MT-2020 networks the international research area have. Any way, one of the main important task in this area is the IoT traffic recognition and prediction. Currently, researchers face a new challenge to their experience, talents and desire to reach new heights in information and communication technologies. The new challenge is IMT-2030 networking technologies and services. In this case, the question with effective IoT traffic prediction methods is still relevant during transition to the next IMT-2030 network and services. IoT it is the ubiquitous conception, on which the new IMT-2030 services also based. For example, Tactile Internet, part of the solutions in digital avatars, and others. There more, these algorithms have to be more efficient and fastly in work with huge data capacity, which characterized the different services of Internet of Things. Recently, there are Machine Learning and Big Data algorithms are took this place of new algorithms for efficient and complex algorithms. In this article, we implement IoT traffic prediction approaches using single step ahead and multistep ahead prediction with NARX neural network. As a data we used the metadata of flows which were received through the northern interface. The prediction accuracy has been evaluated using three neural network traing algorithms: Traincgf, Traincgp, Trainlm, with MSE as performance function in term of using mean absolute percent of error (MAPE) as prediction accuracy measure. IoT