Monitoring and explainable prediction technique for density of hot mix asphalt during pavement construction via intelligent compaction technology

During the construction of Hot Mix Asphalt (HMA) pavements, compaction is crucial for ensuring the pavement's ultimate performance. Traditionally, in-place density is monitored using in situ spot tests with non-destructive density gauges (nuclear or non-nuclear) calibrated against core sample measurements. However, these density spot tests are often conducted at different intervals, hindering timely assessment. Intelligent compaction (IC) has emerged to enhance this process, equipping rollers with sensors to improve HMA compaction. This study aims to monitor the density of the asphalt pavement and develop predictive models for HMA density during construction, utilizing intelligent compaction measurement values (ICMVs), lanes, and pass counts. Data were collected from a construction project in Pakistan, including IC data from vibratory rollers and non-nuclear gauge (NND) density measurements. Machine learning methods—specifically support vector regression, random forest, extreme gradient boosting, and light gradient boosting—were applied and enhanced with a chaotic firefly algorithm to develop these predictive models. The robust model was interpreted using Shapley’s adaptive explanations (SHAP) to assess the influence of ICMVs on NND. The results revealed that the model's predictions closely aligned with measured NND values, with light gradient boosting demonstrating superior performance during validation and achieving an R 2 value of 0.989 and RMSE less than 1.172. The SHAP analysis demonstrated a significant influence of ICMVs, particularly the compaction control value (CCV) and resonant meter value (RMV) with NND. Furthermore, the machine learning model's results outperform previous studies' findings. These findings suggest that utilizing ICMVs and machine learning predictions may effectively replace traditional NND density tests, providing easier and more accurate access for density determination during asphalt construction.