Explainable Ensemble Machine Learning Approach for Predicting the Indirect Tensile Strength of Hot Mix Asphalt

Abstract The primary value of this work is the combination of interpretable machine learning and a big data of pavement to provide accurate prediction and viable input to asphalt mixture. Altogether, the findings indicate that an explainable ensemble machine learning can be adopted as a reliable and understandable method of quick ITS prediction. It is also used to aid in lessening laboratory work and can contribute to an early decision regarding the asphalt mix design. This study aims to predict the value of this property using interpretable machine learning techniques. A dataset consisting of 3,906 data points obtained from the Long-Term Pavement Performance (LTPP) database was utilized. After a rigorous statistical examination and multicollinearity assessment, machine learning models were implemented in the PyCaret environment, and the model with the best performance was selected using the metrics MAE, MSE, RMSE, R 2 , adjusted R 2 , RMSLE, and MAPE. Results showed that the XGBoost model outperformed the others, achieving an R 2 value of 0.9319. Moreover, SHAP (Shapley Additive Explanations) values were used to measure the contribution of associated mix parameters on the model’s predictions. SHAP values help in selecting the optimal design variables efficiently instead of conducting multiple laboratory tests. The primary value of this work is the combination of interpretable machine learning and a big data of pavement to provide accurate prediction and viable input to asphalt mixture. Altogether, the findings indicate that an explainable ensemble machine learning can be adopted as a reliable and understandable method of quick ITS prediction. It is also used to aid in lessening laboratory work and can contribute to an early decision regarding the asphalt mix design.