Time-series short-term forecast of internet traffic

Abstract EN

Internet service providers are expected to continuously provide services without interruptions.

However, planning for this activity requires several tasks (such as forecasting clients’ demand) that need to be considered in fulfilling their objectives particularly in face of calamity.

This paper presents the results of using four methods for forecasting the demand of a single client of Internet Service Provider.

A total of 1440 points of inbound and outbound traffic bandwidth usage, taken at 30 minute time intervals for two weeks, were used.

By visually clustering the traffic and by using different parameters of those forecasting methods, a total of 40 time series combinations were studied.

The results of the study indicate that the best short-term forecasting method was the exponential smoothing method with a smoothing parameter of 0.3. The findings of this study should be of great help to the Internet Service Provider to better plan for load distribution between international links and help in approving down times and maintenance windows requested by cable providers.

This should also help in detecting cyber security attacks or spam and reduce clients’ complaints, thus improving their quality of service. For future work, we recommend to use close parameter values to possibly improve the forecast.