Data supporting the publication: Using Machine Learning to Compute Tire-Penetration Related Properties for Enhanced Rolling Resistance Prediction

This study develops and evaluates an automated, machine-learning-based framework for improving prediction of pavement rolling resistance (RR) using high-resolution pavement surface data from a Laser Crack Measurement System (LCMS). The research focuses on Dutch highway pavement sections, especially PA16/ZOAB-type asphalt surfaces, and investigates how pavement texture, tire-penetration-related properties, and environmental/tire conditions influence the rolling resistance coefficient (RRC). The broader motivation is to reduce uncertainty in transport-related greenhouse gas emissions estimates by improving understanding of tire–pavement interaction.Methodologically, the work combines field data collection, preprocessing, conventional statistical modeling, physical interpretation, and machine learning. A total of 18,782 records were assembled from the Dutch pavement network, including LCMS texture data, rolling-resistance trailer measurements, pavement age/type information, and temperature/tire-pressure variables. LCMS raw cross-profile data were used to calculate texture indicators such as MPD, RMS, ETD, and skewness, and measurement locations were aligned with rolling-resistance observations using GPS and pavement-age matching. Data cleaning, validation, filtering, and standardization were performed before analysis. Rolling resistance was measured at 80 km/h with TU-Gdansk’s rolling-resistance trailer using a Standard Reference Test Tire (SRTT) inflated to 2.1 bar, with tire temperature/pressure, road temperature, and air temperature monitored during testing.The study first reproduces and tests commonly used rolling-resistance prediction models based on simple and multiple linear regression using standard pavement texture indicators. It then applies a varied sampling approach by averaging data over 10 m segments to reduce noise and assess whether prediction performance improves. Next, the authors introduce a tire-penetration analysis workflow based on LCMS depth data and an enveloping technique: a 50 × 50 mm wheel-track patch is extracted, an envelope is applied to the profile, and the volume of deformed rubber and penetration depth are estimated from pixel volumes below a reference surface. This physical feature engineering step is intended to capture tire–pavement interaction more directly than texture indicators alone.For machine learning, the study uses two predictive approaches: an ML-oriented multiple linear regressor and a random forest regressor. Pearson correlation analysis is used for dimensionality reduction and feature selection, identifying redundancy among MPD, ETD, and RMS, and highlighting the importance of variables such as tire temperature, road temperature, air temperature, tire inflation pressure, MPD, and skewness. The dataset is split into training/validation/test sets (60/20/20). Hyperparameters are tuned with RandomizedSearchCV, model robustness is checked with 5-fold cross-validation, and uncertainty is quantified using bootstrap resampling with 100 retrained random-forest models to generate 95% prediction intervals.Key findings are that conventional regression models based only on texture variables perform poorly or only moderately, even when sampling is improved, while the random forest model substantially improves RRC prediction by capturing nonlinear relationships. The analysis also suggests that pavement type and pavement age matter, with older surfaces tending to show higher rolling resistance. The LCMS-based tire-penetration framework successfully estimates penetration depth and deformed-rubber volume, although the study does not find a clear monotonic relationship between penetration-volume estimates and the limited RRC range in this dataset.Regarding data collection governance, the paper describes field measurements on road sections and use of historical/operational pavement data from the Dutch road network, but it does not report human-subject involvement, personal data collection, animal experimentation, or a formal ethics-board approval process. Based on the provided text, the work appears to rely on engineering field measurements and infrastructure data gathered within a collaboration among TU Delft, TNO, Rijkswaterstaat, and the Knowledge-based Pavement Engineering (KPE) research program. No explicit IRB/ethics approval statement is included in the supplied manuscript excerpt.