Traffic predictability based on ARIMA/GARCH model

The predictability of Internet traffic is a significant interest in many domains such as adaptive applications, congestion control, admission control, and network management. In this paper, we propose a new traffic prediction model called autoregressive integrated moving average with generalized autoregressive conditional heteroscedasticity (ARIMA/GARCH), which can capture traffic burstiness and exhibit self-similarity and long-range dependence (LRD). We discuss network traffic predictability related to different prediction applications and measure methods. We validate our prediction model by comparing with other models, includes non-model-based minimum mean square error (MMSE), pure self-similar fractional ARIMA (FARIMA). We use the real network traces to evaluate models. The results show that MMSE computation is simplest and fastest and can apply for online prediction applications. The results also show that FARIMA predictability relies on strong degree of self-similarity, our proposed ARIMA/GARCH model get the best adaptability and accuracy. Therefore ARIMA/GARCH model can be used for exact prediction applications