Flexible Stochastic Frontier Approach to Predict Spot Speed in Two-Lane Highways

Abstract

The approach to spot speed prediction in two-lane highways followed in this study aims to evaluate the effects of a comprehensive set of speed factors, with a special focus on the geometric characteristics of the road segment to which the element belongs. Two flexible models were developed for different types of roads based on a stochastic frontier formulation in which the maximum operating speed is estimated as a function of road geometrics, and the one-sided disturbance accounts for diversity in driving behavior and vehicle characteristics, allowing the estimation of any percentile speed. The models are applicable to horizontal curves and tangents and consider both on-site characteristics and aggregated variables characterizing the road segment. The results show a clear influence of segment features on different percentile spot speeds, revealing that recent driving experience and expectations about the quality of the geometric design influence the way a driver approaches a specific road element. Additionally, this study contributes to addressing some of the limitations of existing speed models identified in the literature.