Application of Artificial Neural Networks to Predict Speeds on Two-Lane Rural Highways

Abstract

The ability to predict accurately vehicular operating speeds is useful for evaluating the planning, design, traffic operations, and safety of roadways. Operating speed profile (OSP) models are used in the geometric design of highways to evaluate design consistency. Design consistency refers to the condition where the geometric alignment does not violate driver expectations. Existing OSP models have been developed using ordinary linear regression methods. However, the assumptions and limitations inherent to linear regression may at the very least complicate model formulation. If not acknowledged and corrected for, deviations from these assumptions can also adversely affect the efficacies of such models. Artificial neural networks (ANNs) are modeling tools that do not impose the stringent assumptions and limitations imposed by regression. It is therefore of interest to know whether ANNs are viable alter natives to linear regression for OSP modeling. Two backpropagation ANNs for operating speed predictions for passenger cars on two-lane rural highways are evaluated, and their performances with regression-based models are compared. The results of these comparisons indicate that the explanatory powers of the ANN models are comparable with those developed by regression. The predictive powers of the two types of models were observed to be comparable, and ANNs were not limited by distributional or other constraints inherent to regression. Therefore, ANNs were determined to be a viable alternative to regression for OSP model construction.