Classification of Driver Behavior in Horizontal Curves of Two-Lane Rural Roads

The components of a road directly affect its safety and, by analyzing the consistency of the geometric design, this condition can be assessed. Vehicle speed is a factor in assessing geometric design consistency, and it is considered constant throughout the driver's transition between the tangents of horizontal curves. However, in most cases, drivers modify the speed depending on the vehicle's position before, during, and after the curve. Therefore, a speed profile allows one to describe this behavior better and study the relationship between speed and the curve elements. Previous literature shows that the rural environmental influences the speed of road users. Additionally, horizontal curve negotiations are linked to a high number of accidents, most of which are caused by driving errors and aggressive/risky driving style. This study aims to determine the optimal number of groups of curve speed profiles that are strongly influenced by the horizontal alignment elements geometry. This research introduces a cluster-based analysis approach to identify speed profiles from field data in two steps; first, the optimal number of speed profiles is determined through the evaluation of several techniques and, second, speed profiles are classified using the k-means method with different sets of data. Free-flow velocity data observed travelling in both directions within 38 horizontal curves on two-lane rural roads in Puerto Rico was used for validation. The validation of the proposed methodology is carried out by directly comparing the number of behaviors observed in users with the geometric and operational characteristics of the curves. The results show that by employing three types of driving styles referred to as Aggressive, Moderate, and Cautious and, adding two inferred variables that consider the time, the driver stays near or above the speed limit and changes acceleration during their movement through the curve, allowing a clearer and more uniform categorization of speed profiles. Furthermore it is observed that the number of grouped profiles have a high dependence on the values taken by the variables of the curve design as expected due to their influence on the perception of safety by drivers.