Answers are grounded in this paper and cite their sources.
Whether daytime condition ratings recorded in traffic sign inventories identify signs failing retroreflectivity requirements is uncertain. This study evaluated the diagnostic performance of an administrative daytime rating in a 2016 Ecuadorian road inventory and whether predictive, explainable machine learning could improve measurement prioritisation. Ratings were compared with retroreflectivity measured on 16,487 plates across 81 road sections, using NEVI-12 in-service minima and accounting for clustering. Blinded photographic ratings, acceptance-sampling simulations, grouped cross-validation, between-province validation and SHAP analysis complemented the evaluation. The inventory rating identified 17.6% of non-compliant plates, while 21.4% of good-rated plates failed. Photographic raters agreed closely but identified only about 31% of failures in the photographic subset. Visually prioritised 10-plate acceptance samples detected at least one failure more often than random samples (0.859 versus 0.770). Among good-rated plates, logistic regression and XGBoost performed similarly (AUC 0.701 and 0.708); SHAP highlighted sign category and installation-related attributes. Performance decreased between provinces. In retrospective simulation, a 5% random sample plus logistic-regression prioritisation identified 55.6% of failures among good-rated plates within the network with a 30% budget. Daytime ratings alone should not determine which individual signs are measured; predictive prioritisation may improve yield, but its performance and operational usefulness require prospective validation.
No takes yet. Share an insight, caveat, or question.
Salazar et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: