Title: Predicting nationwide road fatalities in the US: a neural network approach

Authors: Gokhan Egilmez; Deborah McAvoy

Addresses: Department of Mechanical and Industrial Engineering, University of New Haven, West Haven, CT, USA ' Department of Civil Engineering, Ohio University, Athens, OH, USA

Abstract: Road crashes are among the top five leading causes of deaths in the US although the national trend in fatal crashes has reached to the lowest level since 1949. Therefore, this paper introduces a non-parametric prediction models, artificial neural network (ANN), to assist policy-makers in minimising fatal crashes across the United States. Seven input variables from four safety performance input domains while fatal crash was utilised as the single output variable for the scope of the research. ANN was utilised and the best neural network model was developed out of 1,000 networks. The proposed neural network model predicted data with 84% coefficient of determination. In addition, developed ANN model was benchmarked with a multiple linear regression model and outperformed in all performance metrics including r, R-square and the standard error of estimate.

Keywords: artificial neural networks; highway safety; multivariate regression analysis; prediction; US road fatalities.

DOI: 10.1504/IJMHEUR.2017.10006776

International Journal of Metaheuristics, 2017 Vol.6 No.4, pp.257 - 278

Received: 10 Feb 2016
Accepted: 04 Jan 2017

Published online: 03 Oct 2017 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article