Neural spatial interaction models are receiving much attention in recent years because of their powerful universal approximation properties. They are essentially devices for non-parametric statistical inference, providing an elegant formalism. Neural spatial interaction models have shown considerable successes in a variety of application contexts. The paper discusses a novel modular methodology for neural spatial interaction model identification. We briefly introduce the motivation for the two main constituent components of the methodology: model selection and model adequacy testing. Then we discuss the issues involved in model selection and make a clear distinction between the problems of estimation and model specification. Though major emphasis will be laid on the case of unconstraint spatial interaction, some attention will be paid also to the singly constrained case. The methodology will be illustrated in a real world context. KEYWORDS: Neural Spatial Interaction Models; Model Selection and Model Adequacy Testing; Unconstraint Spatial Interaction.
