Neural Networks for Multidisciplinary Approach Research

The aim of this chapter is to give an overview of the modern world of neural networks to researchers involved in the field of geographical and territorial studies who utilize contemporary approaches that need to be connected to complex analysis of social, economical, environmental, and political data. Classical tools offered by statistical linear analysis reveal increasing weakness, mainly due to two fundamental reasons: the necessity to formulate some a priori hypotheses to determine the optimal tool to apply and the implicit inter-related connections among the variables evolving in a complex system, as an urban territory or a coastal area such as the SECOA case, for instance. For about 40 years, neural networks have been successfully implemented to deal with the complex natural system modelling due to nonlinear nature of connectionist systems and the possibility offered by neural networks to adapt to a huge class of problems, from nonlinear classification to time-series prediction or dimension reduction. The principal neural systems proposed in this chapter are the Feed-Forward Neural Network and the Self-Organized Maps. The first is useful for a nonlinear modelling of the SECOA system, and to the second is better fitted to di similarity analysis (Taxonomy), in order to describe the case studies of conflict territories.