Examining the Feasibility of Using Naturalistic Driving Study Data for Validating Speed Prediction Models

Abstract Numerous curve speed models have been developed for purposes such as predicting driver behavior, evaluating roadway design consistency, and setting curve advisory speeds. With any model development efforts, questions can be raised about the transferability of the model between geographic regions. It is desirable to develop speed prediction models that can be used to predict vehicle speeds across different regions, as these models can then form the basis for consistent roadway analysis and evaluation methods. However, speed model validation and especially calibration are expensive tasks because of the need to collect field data. The main objective of this paper is to present a validation of speed prediction models for horizontal curves using the Naturalistic Driving Study (NDS) database. For this purpose, four two-lane rural highway sections from the State of Indiana were selected where each section includes multiple horizontal curves. Roadway design characteristics were used to predict the speed at the midpoint of each curve based on models that were previously calibrated using Texas data. Actual vehicle speeds at each curve were then computed and compared with the predicted speed using the Texas models. The results of the analysis suggest that the predicted speed from the Texas model provides an unbiased estimate of observed curve midpoint speeds in Indiana, and can provide good estimates of speeds at the beginning and ending of each curve with minor adjustment. The successful use of the NDS database for model validation shows that this database can facilitate similar efforts in additional states.