Network Traffic Prediction Using Online-Sequential Extreme Learning Machine

Abstract

For years, it has been a great challenge for Internet Service Providers (ISP) to predict traffic load or future demand, since each bit of traffic is an economic cost to operators. Additionally, more and more users are adopting the different telecommunication services as well as the ISP must provide more bandwidth and reliability. In this sense, the study focuses on forecasting with neural network techniques, specifically Long Short-Term Memory (LSTM) and Online-Sequential Extreme Learning Machine (OS-ELM), the real traffic of a Chile ISP. The results show that OS-ELM outperforms LSTM in terms of computational cost by a factor of 2300, and in terms of network prediction, OS-ELM effectively competes with LSTM.