In periods of scarce resources, efficient pavement preservation is challenging, as authorities have to ensure pavement serviceability and road safety, under tight budget constraints. Efficient prediction of pavement deterioration is critical for proper decision making  with respect to pavement maintenance and rehabilitation activities. Cracking, in particular, is among those pavement failures, whose repair is of outmost importance for ensuring highway safety and serviceability. As such, different models have been introduced in the literature, attempting to model and predict asphalt pavement cracking.

This paper focuses on comparing the performance of an econometric (Bayesian stochastic duration model - BSD) and a computational intelligence model (neural network - NN) in forecasting cracking probability of asphalt pavements. Both models have been developed using the LTTP dataset and consider various pavement parameters, along with the effect of treatments. A qualitative comparison showed that the NN model involves more explanatory parameters compared to the BSD model; however, common explanatory variables offer similar insights among models. Subsequently, distributions of cracking probability for the two modes were compared using a non-parametric statistical test. Quantitative results revealed that in cases of most treatments, probabilities follow the same distribution with respect to time.

Bayesian model, stochastic duration, neural networks, fatigue cracking, pavements.