A model predicts road roughness with 4.2% average error.

The study’s authors say empirical models miss nonlinear pavement deterioration, especially in data-scarce local networks.

Researchers tested 6 algorithms on 2,313 road segments covering 462.6 km, and LightGBM scored best with an R² of 0.83.

Air voids in the asphalt had the strongest effect on roughness, the study’s interpretability analysis found.

The paper is an accepted early version and will be replaced by a final version of record.

Sources: Nature