By Wei Cai, Cox Communication
Upstream broadband usage and network capacity have been increasing sharply in recent years, particularly under global pandemic lockdowns since March 2020. The spread of COVID19 around the entire world has placed upstream traffic growth on an extremely irregular pattern with fluctuations, posing great challenges to use conventional methods such as Auto-Regressive Integrated Moving Average (ARIMA) and other classical statistical models to ensure accurate forecasts because those classical methods have been proven to be weak and inadequate for modeling non-stationary traffic flows.
In practical forecasting applications, the use of different modeling methods has become a popular research by many scholars. However, with the rapid development of cable companies’ network, network traffic data scale is increasingly important in modern network traffic world. Just combining forecasts from different models with weighs allocated does not satisfy the need to model big traffic flow data more effectively.