A new model for predicting IRI using TabNet with hunter-prey optimization

To address the prediction problem of the international roughness index (IRI), which is influenced by multiple factors, we propose a hunter-prey optimization (H-PO) TabNet prediction model. First, we combined multiple data sources, including traffic flow, climate data, pavement basic, and pavement condition data, aligned with annual detection data for highways at a spatial resolution of 100 m. Second, we conducted feature importance analysis on the integrated dataset to identify the most significant factors influencing pavement surface smoothness, which serve as inputs for the prediction model. We employed multiple machine-learning prediction models to predict the IRI using the compiled multisource dataset. After comparing the predictive performance of different models, we selected the TabNet model as the base model and optimised its parameter tuning process using the H-PO algorithm. Finally, ablation experiments were conducted to validate the proposed model. The results demonstrate that the H-PO-TabNet model achieved the best performance, with an R2 increase of 7.92% compared with a model with default parameter settings. The H-PO-TabNet model also attained the highest overall accuracy, with an R2 value of 0.8763. This model can improve the accuracy of IRI prediction and provide certain data support for pavement maintenance activities.