A comparison of traditional forecasting methods for short-term and long-term prediction of faults in the broadband networks

In this paper we analyze different traditional forecasting methods for prediction of the expected number of faults in broadband telecommunication networks. The dataset consists of over 1 million measured values, collected in recent years. A lot of factors, both in the network and outside the network, contribute to the formation of faults. Therefore, the faults occurring can be considered as a nonlinear time series. A comparison of autoregressive models and conditional heteroscedastic models is presented for short-term and long-term prediction of appearance of faults. Assessment of the accuracy of tested models is made by comparing the results obtained by modeling and the actual data. We are trying to find the best candidates for the analysis and forecasting of faults occurring.