Intelligent monitoring of pavement skid resistance could significantly reduce highway crash rates. However, conventional skid resistance prediction methods often lack mechanistic interpretability. To develop a more explainable friction model based on skid-resistance mechanisms, this study integrated the “skid-resistance contribution zone” (SRCZ) concept with a light gradient boosting machine (LightGBM) framework. The experiment program employed a scanning laser to acquire 105 three-dimensional profiles of a series of pavement with different textures. Through tire-pavement contact testing and digital image processing, the effective contact texture area was determined, enabling the SRCZ extraction from the full-depth texture. Twelve parameters were extracted to quantify both full-depth textures and SRCZ features. Following pavement friction measurements using a British pendulum tester, LightGBM was employed to establish the prediction model. Results indicated that the SRCZ-enhanced LightGBM model achieved superior performance (R 2 = 0.91) compared with conventional full-depth parameter approaches. Feature importance analysis revealed that the mean volume of isolated island exhibited the strongest predictive contribution. Comparative analysis also demonstrated that LightGBM has superior predictive capability over five alternative machine learning algorithms.
Chen et al. (Tue,) studied this question.