This paper compares four different artificial neural network approaches for computer network traffic forecast, such as: 1) multilayer perceptron (MLP) using the backpropagation as training algorithm; 2) MLP with resilient backpropagation (Rprop); (3) recurrent neural network (RNN); 4) deep learning stacked autoencoder (SAE). The computer network traffic is sampled from the traffic of the network devices that are connected to the internet. It is shown herein how a simpler neural network model, such as the RNN and MLP, can work even better than a more complex model, such as the SAE. Internet traffic prediction is an important task for many applications, such as adaptive applications, congestion control, admission control, anomaly detection and bandwidth allocation. In addition, efficient methods of resource management, such as the bandwidth, can be used to gain performance and reduce costs, improving the quality of service (QoS). The popularity of the newest deep learning methods have been increasing in several areas, but there is a lack of studies concerning time series prediction, such as internet traffic.

Abstract. Internet traffic prediction is an important task for many applications, such as adaptive applications, congestion control, admission control, anomaly detection and bandwidth allocation. In addition, efficient methods of resource management can be used to gain performance and reduce costs. The popularity of the newest deep learning methods has been increasing in several areas, but there is a lack of studies concerning time series prediction. This paper compares two different artificial neural network approaches for the Internet traffic forecast. One is a Multilayer Perceptron (MLP) and the other is a deep learning Stacked Autoencoder (SAE). It is shown herein how a simpler neural network model, such as the MLP, can work even better than a more complex model, such as the SAE, for Internet traffic prediction.

Resumo: Neste artigo são comparados quatro abordagens diferentes para a previsão do tráfego de redes de computadores, usando o tráfego de dispositivos de redes de computadores que se conectam a Internet e usando Redes Neurais Artificiais (RNA) para a predição, sendo elas: (1) Multilayer Perceptron (MLP) com Backpropagation para o treinamento; (2) MLP com Resilient Backpropagation (Rprop); (3) Rede Neural Recorrente (RNN); (4) Stacked Autoencoder (SAE) com aprendizagem profunda (deep learning). Também é apresentado que um modelo de rede neural mais simples, tais como a RNN e MLP, podem ser mais eficientes do que modelos mais complexos, como o SAE. A predição do tráfego de Internet é uma tarefa importante para muitas aplicações, tais como aplicações adaptativas, controle de congestionamento, controle de admissão, detecção de anomalias e alocação de largura de banda. Além disso, mé-todos eficientes de gerenciamento de recursos, como a largura de banda, podem ser usados para melhorar o desempenho e reduzir custos, aprimorando a Qualidade de Serviço (QoS). A popularidade das novas redes neurais profundas vêm aumentado em muitas áreas, porém há uma falta de estudos em relação a predição de séries temporais, como o tráfego de Internet.