Using AI to Predict Permeability of Porous Asphalt Pavements

Abstract

Porous asphalt pavements play an important role in stormwater management while also providing structural integrity. Reliable permeability prediction is dynamic, allowing for better design optimization, long-term functioning evaluation, and effective maintenance planning. This work creates and evaluates artificial intelligence (AI) models for predicting the hydraulic permeability of porous asphalt pavements using a multi-regional dataset comprising 1,250 laboratory and field observations. The information included mix design parameters (air voids, D10/D50 ratio, binder concentration), compaction, construction, climate, and maintenance history. Machine-learning techniques such as artificial neural networks (ANN), support vector machines (SVM), random forest, and gradient boosting were trained and tested using 10-fold cross-validation and temporal validation (2010–2017 training; 2018–2021). ANN provided the greatest performance (R² = 0.91, RMSE ≈ 0.11 cm/s) with a