Citation

Chimba D, Sando T. Adv. Transp. Stud. 2009; 19: 17-26.

Affiliation

Stanley Consultants, 1601 Belvedere Rd, Suite 400 E, West Palm Beach, FL, 33406 (chimbadeo@stanleygroup.com); School of Engineering University of North Florida, 1 UNF Drive, Jacksonville, FL 32224 (t.sando@unf.edu).

Copyright

(Copyright © 2009, Arcane Publishers)

DOI

unavailable

PMID

unavailable

Abstract

This paper describes the use of one of the neuromorphic techniques – Artificial Neural Networks (ANN) Backpropagation technique to predict crash injury severity. The method of optimizing the number of neurons and epochs used in the ANN backpropagation architecture is presented. The paper also compared the accuracy of the backpropagation method with that of the Ordered Probit (OP) model. The prediction accuracies of 83.3% and 65.5% were obtained for the ANN backpropagation and Ordered Probit (OP) models, respectively. The results indicate that a well structured network with optimized number of neurons and epochs, ANN can perform better than a traditional OP technique. It was also noted that the choice of the number of epochs and neurons is key to obtain an efficient ANN architecture.