Artificial Neural Networks: A New Approach to Modelling Interregional Telecommunication Flows

摘要: This paper suggests a new modelling approach, based upon general nested sigmoid neural network model. Its feasibility is illustrated in the context of interregional telecommunication traffic Austria and its performance evaluated comparison with classical regression approach gravity type. The application this may be viewed as three-stage process. first stage refers to identification an appropriate from family two-layered feedforward networks three input nodes, one layer (sigmoidal) intermediate nodes output node. There no procedure address problem. We solved issue experimentally. input-output dimensions have been chosen order make model close possible. second involves estimation parameters selected performed via adaptive setting (training, estimation) by means least mean squared error goal back-propagating technique, recursive learning using gradient search minimise goal. Particular emphasis laid on sensitivity choice initial well problem overfitting. final applying testing teletraffic flows predicted. Prediction quality analysed two measures, average relative variance coefficient determination, use residual analysis. analysis shows that outperforms Austria.