Analysis and Categorization of Traffic Streams by Artificial Intelligence

Abstract

This report presents an evaluation of artificial neural networks in terms of computational efficiency, by analyzing transmitted information flows for determination the type of defined traffic categories using artificial intelligence. The subject of study are Markov M/M/c circuits with unlimited number of waiting calls in the queue and fixed number of server stations in accordance with the desired test categories, as follows c=5, c=10 and c=15. Three layer architectures are applied to different types of neural output activators with Levenberg-Marquardt training, respectively linear, tangent-sigmoidal and logarithmic-sigmoidal. The lowest values of the Mean Squared Error (MSE) of 0.0080, 0.0041, and 0.1923 are experimentally established at 7, 3, and 25 hidden neurons for the indicated activation functions. An accuracy levels of 94.4%, 100.0%, and 70.6% were obtained against indicator levels for identical numbers of neurons.