ARTIFICIAL NEURAL NETWORKS - A NEW APPROACH TO MODELING INTERREGIONALTELECOMMUNICATION FLOWS

During the last thirty years there has been much research effort in re gional science devoted to modeling interactions over geographic space. Theoretical approaches for studying these phenomena have been modifie d considerably. This paper suggests a new modeling approach, based upo n a general nested sigmoid neural network model. Its feasibility is il lustrated in the context of modeling interregional telecommunication t raffic in Austria, and its performance is evaluated in comparison with the classical regression approach of the gravity type. The applicatio n of this neural network approach may be viewed as a three-stage proce ss. The first stage refers to the identification of an appropriate net work from the family of two-layered feedforward networks with 3 input nodes, one layer of (sigmoidal) intermediate nodes and one (sigmoidal) output node (logistic activation function). There is no general proce dure to address this problem. We solved this issue experimentally. The input-output dimensions have been chosen in order to make the compari son with the gravity model as close as possible. The second stage invo lves the estimation of the network parameters of the selected neural n etwork model. This is performed via the adaptive setting of the networ k parameters (training, estimation) by means of the application of a l east mean squared error goal and the error back propagating technique, a recursive learning procedure using a gradient search to minimize th e error goal. Particular emphasis is laid on the sensitivity of the ne twork performance to the choice of the initial network parameters, as well as on the problem of overfitting. The final stage of applying the neural network approach refers to the testing of the interregional te letraffic flows predicted. Prediction quality is analyzed by means of two performance measures, average relative variance and the coefficien t of determination, as well as by the use of residual analysis. The an alysis shows that the neural network model approach outperforms the cl assical regression approach to modeling telecommunication traffic in A ustria.