Explainable Machine Learning-Based Prediction of Pavement Deflection Incorporating Traffic Loading, Cracking, and Structural Characteristics

Pavement deflection, commonly measured using deflectograph devices such as the Lacroix deflectograph, is a key indicator of structural capacity and load-bearing performance of flexible pavements. Accurate prediction of deflection is essential for assessing pavement health and optimizing maintenance and rehabilitation strategies, particularly under semi-arid climatic conditions.

This study proposes an explainable machine learning framework to predict pavement deflection based on multi-source data, including pavement age, cumulative traffic loading (ESAL), cracking severity, subgrade characteristics, structural properties, and climatic conditions. The dataset consists of measurements collected along a 50 km flexible pavement section over the period 1992–2018 in a semi-arid environment.

Several predictive models were developed and compared, including Multiple Linear Regression (MLR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost). The models were further employed to predict pavement deflection over a future horizon of 10 to 15 years. Model validation was carried out by comparing predicted and observed values, using standard statistical indicators such as the coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE).

The results show that XGBoost significantly outperforms conventional regression and ensemble approaches, achieving high predictive accuracy with R² values exceeding 0.90 and low RMSE, indicating strong agreement between predicted and measured deflection values. The inclusion of cracking severity and traffic loading proved critical for accurately capturing the structural response of pavements under repeated loading. In addition, SHAP (SHapley Additive exPlanations) analysis was used to enhance model interpretability and identify the most influential parameters governing pavement deflection.

The proposed approach provides a reliable and interpretable tool for predicting pavement structural performance, supporting data-driven decision-making in pavement management systems and enabling optimized maintenance and rehabilitation planning in semi-arid regions.