Soft Computing, Parametric and Non-Parametric Statistics:A Review and Evaluation

The present paper compares - from a conceptual and modeling perspective - Neural Networks (NNs) and statistical models in transportation and traffic engineering applications. The differences, similarities and topological analogies of both methodological approaches are discussed. Moreover, the state-of-practice in modeling with NNs is reviewed with focus on the comparative studies with classical statistical modeling. The paper also provides a set of guiding arguments on the optimal selection of modeling approach, as well as for efficient design, learning and evaluation of NNs. Possible interactions and collaboration of NNs and classical statistics are also discussed.