Neural Network Modeling of 85th Percentile Speed for Two-Lane Rural Highways

Abstract

The study was undertaken to develop neural network models to predict 85th percentile speed for two-lane rural highways in Oklahoma. Several input parameters, namely, physical characteristics of road, traffic parameters, pavement condition indices, and accident data, were considered in developing the neural network models. The physical characteristics of road include surface width, shoulder type, and shoulder width. The traffic parameters cover average daily traffic (ADT) and posted speed. The pavement condition parameters include skid number and international roughness index (IRI). The location and statewide collision rates (overall, fatal, and injury), and percentage of drivers traveling at an unsafe speed were covered in the accident data. Four models were developed. Model 1 included physical characteristics of road and traffic parameters, including posted speed. Model 2 covered all parameters in Model 1 except posted speed. Similarly, Model 3 considered accident data with all parameters in Model 1. Model 4 used all the parameters in Model 3 except posted speed. Models 1 and 3 were more accurate than Models 2 and 4. The parametric study of the model parameters showed that 85th percentile speed decreased with an increase in the accident rate, ADT, skid number, and IRI. Similarly, widening of the surface and shoulder resulted in a higher 85th percentile speed. It is expected that developed neural network models would be an effective tool for the Oklahoma Department of Transportation and other departments of transportation to enhance traffic safety.