To address the constraints of conventional evaluation methods for asphalt pavement anti-skidding performance, such as insufficient generalization across gradation scenarios and difficulties in collaborative quantification of multi-scale features, this study proposes a composite parameter-based asphalt pavement anti-skidding performance evaluation model using the XGBoost algorithm. The study selected three typical types of asphalt pavements, AC13, AC20, and SMA13, which are widely used in paving practice. The texture feature information of these test road sections was extracted and developed a full-scale texture characterization system by integrating the statistical features (energy and entropy) and fractal features (fractal dimension and multifractal spectrum width Δα). The feature segmentation counting method was employed to quantify the contribution mechanism of different texture features to skid resistance, and the application effectiveness of the model in cross-gradation scenarios was systematically evaluated. The results showed that XGBoost-based composite parameter asphalt pavement anti-skidding performance evaluation model achieved test set accuracy of R² > 0.9 and RMSE < 0.06, confirming its capability to accurately and effectively assess asphalt pavement anti-skidding performance. Model performance was influenced by pavement gradation type: AC13 fine-graded mixtures yielded optimal prediction, SMA13 mixtures ranked second, while AC20 coarse-graded mixtures exhibited relatively weaker generalization but still met requirements. Feature importance analysis identified fractal dimension as the core predictive factor of the model, with different gradation types significantly affecting the effectiveness of texture feature characterization. The research provides significant theoretical and technical support for achieving intelligent detection and gradation-adaptive evaluation of asphalt pavement anti-skidding performance.