Influence evaluation and prediction of the fatigue life of calcium sulfate whisker-modified asphalt binder based on machine learning

Whisker-modified asphalt offers better fatigue and thermal performance for pavements. Laboratory tests for evaluating and predicting the fatigue property of whisker-modified asphalt binder are time-consuming and labor-intensive. Therefore, it is necessary to study a feasible approach for the scientific prediction of the fatigue property of whisker-modified asphalt binder. In this study, 324 fatigue life values of calcium sulfate whisker-modified asphalt binders, prepared by three types of asphalt binders and three aging conditions and four dosages of calcium sulfate whiskers, were obtained based on the linear amplitude sweep test. The multiple linear regression (MLR) model and three machine learning models—random forest model, gradient boosting regression (GBR) model, and XGBoost model—were compared, and three indicators—determination coefficient, root-mean-square error, and mean absolute percentage error—were adopted to evaluate the suitability and accuracy of models on the prediction the fatigue life of calcium sulfate whisker-modified asphalt binders. The fatigue life of asphalt binders was significantly influenced by the asphalt types and ultraviolet aging, and there was no significant correlation between whisker contents and the fatigue life of asphalt binders. Three machine learning models significantly outperformed the MLR model in the prediction accuracy of the fatigue life of asphalt binders. The XGBoost model had the largest determination coefficient values (0.9998) and the smallest values of mean absolute percentage error (0.15%), which corresponds to a 97% reduction in prediction error compared to the MLR model. GBR model showed the most stable prediction of fatigue life for CSW-modified binders.