Prediction of Pavement Condition Rating (PCR) Using Interpretable Machine Learning Framework: Evidence From Saudi Arabia

Effective pavement monitoring and maintenance is essential for safe, durable and well performing road network, especially in harsh climate regions like Saudi Arabia. Conventional mechanistic and deterministic approaches fail to capture the complex pavement deterioration behavior under influence of heavy traffic and severe environmental conditions. This study presents a robust integrated machine learning based framework for the prediction of Pavement Condition Rating (PCR) using pavement condition and weather data from rural highways in Saudi Arabia. This proposed PCR modelling framework aims to compare the prediction performance of Light-GBM and Cat-Boost against baseline Support Vector Machine and Random Forest models. To enhance model transparency and interpretability, this research also aims to conduct feature importance analysis and SHAP analysis to identify the significance of input predictors. Models' assessment showed that both Cat-Boost and Light-GBM achieved a prediction accuracy of 0.87 surpassing the two conventional models. Feature importance analysis results revealed that traffic loading (AADT) emerges as a most influential parameter on PCR followed by pavement distress related such as rutting, cracking and temperature. The practiced methodological approach provides an actional framework for local practitioners and other similar jurisdictions for proactive identification and prioritization of pavement maintenance activities.