Description

Forecasting road condition after maintenance can help in better road maintenance planning. As road administrations annually collect and store road-related data, data-driven methods can be used in determining forecasting models that result in improved accuracy. In this paper, we compare the prediction models identi fied by experts and currently used in road administration with simple data-driven prediction models, and parsimonious models based on a input selection algorithm. Furthermore, non-linear prediction using radial basis function networks is performed. We estimate and validate the prediction models with a database containing data of over two million road segments.

Forecasting road condition after maintenance can help in better road maintenance planning. As road administrations annually collect and store road-related data, data-driven methods can be used in determining forecasting models that result in improved accuracy. In this paper, we compare the prediction models identi fied by experts and currently used in road administration with simple data-driven prediction models, and parsimonious models based on a input selection algorithm. Furthermore, non-linear prediction using radial basis function networks is performed. We estimate and validate the prediction models with a database containing data of over two million road segments.