Distance-Based Speed Prediction Models Using Naturalistic Driving Data on Two-Lane Roads in Mountainous Regions

With the advent of transportation, highway crashes are a major problem affecting human health and the environment. In which, a driver's behavior is the most significant factor affecting highway crashes. Because the speed the driver selects depends on their perception of the curve geometry, it is necessary to identify the effects of curve geometry factors on the driver's speed choice behavior. The present study attempts to develop a distance-based speed prediction model with varying curve radius and grade. For that, 64 curves were selected with the variable curve geometry, especially curve radius. The speed data were collected using the performance box (P-Box) for cars and heavy commercial vehicles (HCVs). A total of 24 P-Box runs were carried out for the subject vehicles. After that, eight different machine learning regression techniques were used to develop the distance-based speed prediction model. Out of them, the AdaBoost regression was found to be more accurate than the other machine learning regression technique. Further, the model results revealed that for similar curve geometry, the drivers’ speed choice behavior and other vehicle characteristics significantly influence car and HCVs’ speed. Based on the study outcome, highway authorities can provide some proactive measures to reduce the percentage of curve negotiation and increase the operational efficiency of the highway alignments.