Wireless airtime traffic estimation using a state space model

A new forecasting technique called the extended structural model (ESM) is presented. This technique is derived from the basic structural model (BSM) by the introduction of extra parameters that were assumed to be 1 in the BSM. The ESM model is constructed from the training sequence using the standard Kalman filter recursions, and then the extra parameters are estimated to minimize the mean absolute percentage error (MAPE) of the validation sequence. The model is evaluated by prediction of the total number of minutes of wireless airtime per month on the Bell Canada network. The ESM model shows an improvement in MAPE of the test sequence over both the BSM and seasonal autoregressive integrated moving average. The improved prediction can significantly reduce the cost for wireless service providers, who need to accurately predict future wireless spectrum requirements