Desarrollo de un algoritmo de redes neuronales artificiales aplicado a la predicción de tráfico de la infraestructura de comunicaciones de redes corporativas.

Abstract

An algorithm for prediction of network traffic based on artificial neural networks in corporate networks was developed. In order to get the traffic data generated by the network of Computer Science and Electronics School (FIE), a data collection server with Linux was installed, using TCPdump and a script to capture automatically data during 4 weeks between 8am - 8pm, with intervals of 5 minutes. Trafic data collected contain the information of selected protocols due to they occupy more space in each package. Internet Protocol (IP), Internet Protocol Version 6 (IPV6), Address Resolution Protocol (ARP), Hypertext Transfer Protocol (HTTP), Hypertext Transfer Protocol Secure (HTTPS), and ISCSI-TARGET. The size of each protocol involved in every scan through the Program SteelCentral Packet Analyzer for the storage of the data was analyzed. For the data storage Excel was used for the creation of a database and then export the data to Matlab where the Artificial Neural Network with a backpropagation algorithm for the prediction of data was performed. A sample of 2880 data was collected, analyzed, entered, and processed in the algorithm which consists of 300 entries, 1 hidden layer that consists of 17 neurons, and a single output. All data mentioned above were used for training the artificial neural network. It is concluded that the Artificial Neural Network with a backpropagation algorithm after the tests with different numbers of neurons obtained for each of the protocols the following errors, IP protocol 9.50%, IPv6 15.38%, ARP 10.24%, HTTP 15.89%, HTTPS 12.12%, and ISCSI- TARGET 6.22%. For the traffic prediction of corporate networks, it is recommended to make an artificial neural network whose characteristics are greater than 300 input neurons, a hidden layer with 17 or more intermediate neurons.