Abstract
Artificial intelligence (AI) is applied to predict pavement condition, which supports maintenance decisions and reduces resource use. In this study, an ML model is developed for pavement condition evaluation, which relies on factors such as asphalt temperature and pavement thickness. Field data from Heavy Weight Deflectometer (HWD) and Ground Penetrating Radar (GPR) tests are used to confirm model accuracy. Among the evaluated approaches, the Gated Recurrent Unit (GRU) model is identified as the most effective, because higher accuracy and lower error are obtained. The results indicate that deep neural based assessment supports early maintenance planning. As a result, repair costs are reduced and road safety is improved. In addition, support is provided for smarter, more resilient, and more sustainable cities. Both GRU and LSTM models are shown to deliver reliable predictions. However, GRU performance is slightly superior, as lower RMSE and higher correlation are achieved. Temporal patterns are effectively captured by GRU, even when unseen data are applied. LSTM performance is also considered satisfactory. However, higher complexity and slightly lower accuracy are observed. Overall, a better balance between accuracy and efficiency for pavement condition prediction is achieved by the GRU model.