Title: A comparative study of M5P, ANN and RENB models for prediction of vulnerable road accident frequency

Authors: Saurabh Jaglan; Praveen Aggarwal; Sunita Kumari

Addresses: Department of Civil Engineering, Deenbandhu Chhotu Ram University of Science and Technology, Sonipat, Haryana, India ' Civil Engineering Department, National Institute of Technology (Kurukshetra), Kurukshetra, Haryana, India ' Department of Civil Engineering, Deenbandhu Chhotu Ram University of Science and Technology, Sonipat, Haryana, India

Abstract: The present investigation aims to evaluate the performance of different models to calculate the Vulnerable Road User Accident Frequency (VRUAF). Nine road stretches were chosen to measure the road geometry and other similar characteristics. Many field studies were conducted to gather information about road geometry, traffic surveys and accident characteristics. A total of 17 input parameters were collected for accident frequency analysis, and three prediction approaches were applied: Fixed/Random Effect Negative Binomial (FENB/RENB) regression models, Artificial Neural Network (ANN) and M5P model tree. The variation in models' performance was observed in terms of the coefficient of correlation (0.943-0.981), root mean square error (2.274-1.655) and mean absolute error (1.746-1.351). The result suggests that the ANN model is the most accurate model where CC, MAE and RMSE values are 0.981, 1.351 and 1.655, respectively. Thus, this model can synthetically predict VRUAF under similar geometric conditions.

Keywords: vulnerable road user; accident frequency; artificial neural network; random effect negative binomial model; M5P model tree.

DOI: 10.1504/IJRS.2024.140610

International Journal of Reliability and Safety, 2024 Vol.18 No.3, pp.231 - 251

Received: 03 May 2023
Accepted: 14 Nov 2023
Published online: 28 Aug 2024 *

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