Pavement Friction Prediction Based Upon Multi-View Fractal and the XGBoost Framework

Abstract

The anti-slip performance of road surfaces directly affects traffic safety, yet existing evaluation methods based on texture features often suffer from limited interpretability and low accuracy. To overcome these limitations, a portable 3D laser surface analyzer was used to acquire road texture data, while a dynamic friction coefficient tester provided friction measurements. A multi-view fractal dimension index was developed to comprehensively describe the complexity of texture across spatial, cross-sectional, and depth dimensions. Combined with road surface temperature, this index was integrated into an XGBoost-based prediction model to evaluate friction at driving speeds of 10 km/h and 70 km/h. Comparative analysis with linear regression, decision tree, support vector machine, random forest, and backpropagation (BP) neural network models confirmed the superior predictive performance of the proposed approach. The model achieved backpropagation (R2) values of 0.80 and 0.82, with root mean square errors (RMSEs) of 0.05 and 0.04, respectively. Feature importance analysis indicated that fractal characteristics from multiple texture perspectives, together with temperature, significantly influence anti-slip performance. The results demonstrate the feasibility of using non-contact texture-based methods to replace traditional contact-based friction testing. Compared with traditional statistical indices and alternative machine learning algorithms, the proposed model achieved improvements in R2 (up to 0.82) and reduced RMSE (as low as 0.04). This study provides a robust indicator system and predictive model to advance road surface safety assessment technologies.