Hybrid Modeling of the Luminance Coefficient of Bituminous Mixtures Using a Generalized Additive Model and Data Mining Methods

The paper introduces a non-linear method for modeling the luminance coefficient (Qd) of asphalt (bituminous) mixtures using a Generalized Additive Model (GAM). Developed from observations after three and six months of service, the model accounts for the effects of aggregate luminance, binder content, and air voids, as well as temporal and non-linear dependencies. It showed a high goodness-of-fit (R2 = 0.91) and strong predictive accuracy (RMSE = 4.8 mcd/m2/lx). The analysis revealed that the service period significantly influences luminance, with values after six months being, on average, 12.6 mcd/m2/lx higher than at three months. The impact of aggregate luminance was non-linear, displaying a saturation effect, while asphalt content and air voids varied in their influence over time. Results indicate that the factors affecting bituminous mixture luminance are complex and vary over time; moreover, high aggregate luminance alone does not guarantee a high Qd. Applying the additive model confirms the importance of accounting for non-linear effects and temporal interactions when assessing road surface optical properties.