This paper deals with the correlation among topological accessibility of transport networks and the spatial interaction of activities. Neural networks (ANNs) were used in this case study to capture the structure of the relations between transport and land use, and to find the configuration of weights that produces accessibility results similar to those calculated by other widely used mathematical methods (sum of power series method, matrix eigenvalues and principal eigenvector etc.) or multivariate techniques (cluster, regression, structural equations model). The findings show that ANNs produce in 96-99% of cases the same ranking lists as do matrix algebra techniques. We can state that ANNs represent a competitive candidate for developing an internal representation of topological data, demonstrating that accessibility characterises a well- defined structure, with clear relationships between data, and ANNs do well in modelling it. Moreover, ANNs have shown good modelling power when applied to a 'noisier' data set of accessibility measures. (a) For the covering entry of this conference, please see ITRD abstract no. E206301.
Samenvatting