Exploring pavement friction variability factors using ensemble trees and causal inference

Understanding pavement friction measurement data is necessary to predict future road conditions and determine intervention strategies. Although considerable friction measurement data are collected for these purposes, it is not yet entirely clear how to interpret it. More specifically, there is often unexplained variability associated with these data, which inhibits their use. In this study, we have focused on enhancing the understandability of the data by exploring the causes of the unexplained variability. We constructed a dataset from two decades of friction data on Swiss national roads to explore the influence of different factors, including systematic testing conditions and external factors, on the observed data variations. We used average difference to quantify the degree of variability between consecutive measurements. Explainable ensemble trees and the SHapley Additive exPlanations methods are applied to assess the factors’ contribution to the data variability. Furthermore, a structural causal framework is employed to unravel the factors’ causal effects. Our findings indicate that much of the unexplained variability is related to maintenance interventions, temperature differences, and the speed at which the measurements were taken. These findings demonstrate how the data mining methods confirm the patterns observed in measurements conducted in controlled experiments.