Since the 2000s, numerous studies have developed models to predict the decay of lane marking quality, incorporating variables such as marking material and age, traffic volume, pavement surface, climate conditions, and winter maintenance. These models exhibit significant variability underscoring the need for models tailored to different regions and local practices. A dataset was collected encompassing road geometry, traffic volumes, and weather conditions from approximately 250 km of two-lane rural roads over three years of retro-reflectivity (RL) field data. By the comparison of different Machine Learning (Random Forest, XGBoost) and statistical (GEE) tools, XGBoost emerged as the best model due to its superior prediction accuracy and computational efficiency. In defining a more parsimonious model, we focused on significant variables identified from the XGBoost and GEE models that showed less relevance in the model performance and are easily accessible to road agencies. Both parsimonious models demonstrated performance metrics comparable to their comprehensive counterparts, confirming their suitability for making accurate predictions. Specifically, the parsimonious XGBoost model achieved an R² value of 0.67 compared the value of 0.52 for the complete GEE model, highlighting its effectiveness in practical applications. An efficiency analysis showed that the XGBoost parsimonious model is well-suited for road maintenance, providing reliable prediction curves and valuable insights for planning preventive maintenance actions and reducing survey costs.

Development of decay curve for paint pavement marking using machine learning and generalized estimation equation

Dimauro, Giovanni Andrea
Primo
;
Cafiso, Salvatore
Secondo
;
Ghaderi, Omid
Penultimo
;
Pappalardo, Giuseppina
Ultimo
2025-01-01

Abstract

Since the 2000s, numerous studies have developed models to predict the decay of lane marking quality, incorporating variables such as marking material and age, traffic volume, pavement surface, climate conditions, and winter maintenance. These models exhibit significant variability underscoring the need for models tailored to different regions and local practices. A dataset was collected encompassing road geometry, traffic volumes, and weather conditions from approximately 250 km of two-lane rural roads over three years of retro-reflectivity (RL) field data. By the comparison of different Machine Learning (Random Forest, XGBoost) and statistical (GEE) tools, XGBoost emerged as the best model due to its superior prediction accuracy and computational efficiency. In defining a more parsimonious model, we focused on significant variables identified from the XGBoost and GEE models that showed less relevance in the model performance and are easily accessible to road agencies. Both parsimonious models demonstrated performance metrics comparable to their comprehensive counterparts, confirming their suitability for making accurate predictions. Specifically, the parsimonious XGBoost model achieved an R² value of 0.67 compared the value of 0.52 for the complete GEE model, highlighting its effectiveness in practical applications. An efficiency analysis showed that the XGBoost parsimonious model is well-suited for road maintenance, providing reliable prediction curves and valuable insights for planning preventive maintenance actions and reducing survey costs.
2025
Generalized estimation equation (GEE)
Machine learning (ML)
Preventive maintenance
Retroreflected luminance (RL)
Retroreflectivity degradation models
Road markings
Road safety
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.11769/669749
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? 0
social impact