Abstract
This study develops data-driven Pavement Condition Index (PCI) forecasting for rigid (jointed concrete) airfields to support maintenance planning, risk mitigation, and lifecycle analysis. Traditional PCI surveys per ASTM D5340 are manual, time-consuming, and subjective, motivating machine-learning alternatives. Using panel-level observations, five models predict next-year PCI from readily available variables: prior-year PCI, annual aircraft traffic, temperature, and rainfall. The models include ordinary least squares (linear), ridge, lasso, random forest (RF), and a feedforward artificial neural network (ANN). All models performed well on held-out validation data; the ANN was best, achieving ��2 = 0.985, MSE = 0.286, MAE = 0.397, and RMSE = 0.534. The ANN architecture comprised one input layer, three hidden layers (128, 64, 32 neurons), and one output layer, with diagnostics indicating no material overfitting. Despite limited data, results show strong potential for accurate short-term PCI prediction in rigid airfields. Main limitations stem from the small sample size and potential generalisability issues. Future work should expand datasets, add predictors (e.g., soil type, traffic mix, slab age, joint spacing, deflection data), and assess ensemble and probabilistic models. The approach can help authorities estimate short-term deterioration and prioritise budgets for timely interventions.
Keywords
airfield, concrete pavement, PCI, data driven, machine learning.
DOI
10.5703/1288284318144
Recommended Citation
Shehu, Abubakar Ahmad; Mohamed, Mohamed Shadi; Zakwan, Arab M.; and Jiang, Jinhui, "Machine-Learning PCI Forecasts for Concrete Pavements: Evidence from a Nigerian Airfield" (2025). International Conference on Durability of Concrete Structures. 1.
https://docs.lib.purdue.edu/icdcs/2025/slm/1
Machine-Learning PCI Forecasts for Concrete Pavements: Evidence from a Nigerian Airfield
This study develops data-driven Pavement Condition Index (PCI) forecasting for rigid (jointed concrete) airfields to support maintenance planning, risk mitigation, and lifecycle analysis. Traditional PCI surveys per ASTM D5340 are manual, time-consuming, and subjective, motivating machine-learning alternatives. Using panel-level observations, five models predict next-year PCI from readily available variables: prior-year PCI, annual aircraft traffic, temperature, and rainfall. The models include ordinary least squares (linear), ridge, lasso, random forest (RF), and a feedforward artificial neural network (ANN). All models performed well on held-out validation data; the ANN was best, achieving ��2 = 0.985, MSE = 0.286, MAE = 0.397, and RMSE = 0.534. The ANN architecture comprised one input layer, three hidden layers (128, 64, 32 neurons), and one output layer, with diagnostics indicating no material overfitting. Despite limited data, results show strong potential for accurate short-term PCI prediction in rigid airfields. Main limitations stem from the small sample size and potential generalisability issues. Future work should expand datasets, add predictors (e.g., soil type, traffic mix, slab age, joint spacing, deflection data), and assess ensemble and probabilistic models. The approach can help authorities estimate short-term deterioration and prioritise budgets for timely interventions.
