Forecasting the future is the prime motivation behind the search for laws that explain certain phenomena. Timeseries analysis is an important statistical tool to study the behavior of time dependent data and forecast future values depneding on the history of variations in the data. Many available techniques for timeseries analysis assume linear relationships among variables. But in the real world, temporal variations in data do not exhibit simple regularities and are difficult to analyze and predict accurately. One of the methods to solve this problem is neural networks. Neural networks have been widely studied and applied to variety of areas. And neural networks are able to perform non-linear modeling and adaption. Hence it is more robust and better in the case of noisy timeseries data.
The problem of obtaining accurate forecasts of the telecommunication demand is of major importance to the telephone industry, since the forecasts are the fundamental inputs to both the short and the long-term planning that takes place in the individual companies. The month-to-month changes in both the telephone subscribing demand and celluar service demand are particularly important to manpower planning. The cost and the quality of the installation service depend on accurate forecasts.
In this study, various methods to forecast the telecommunication demand are performed, and the comparison results of the methods are also showed. The results in this study summarize as follows.
First, in the telephone subscribing demand forecasting, timeseries model, neural networks models, and integrated models are tested to compare the performance of the forecasting accuracy. The result is that NN models are more accurate than timeseries model, and integrated models are slightly better than NN models.
Second, in the cellular service demand forecasting, diffusion model, Gompertz model, and NN models are compared. The results of the performance are the same as those of the telephone subscribing demand forecasting cases. NN models show better performance than diffusion model and Gompertz model.
