On the Transferability of Speed Prediction Models: A Case for Cluster-Modeling on Large Road Networks
摘要
Operating speed is influenced by several factors related to road geometry, roadway environment, vehicle, and driver. Over years, several operating speed prediction models have been developed across the globe. Researchers generally regard operating speed prediction model as non-transferable, owing to variability in unobserved factors, such as driver behavior. This paper focuses on performing a transferability check on operating speed prediction models. The transferability is evaluated based on consensus of speed prediction models when applied on a carefully generated hypothetical network that confirms to site selection criteria of selected models. As expected, a statistical check confirms the lack of consensus across models, and, therefore, the non-transferability. The range of speed predicted for the hypothetical network implied the non-transferability of the speed model—not only in the case of models from distant geographies but even for models developed in the same region. Subsequently, we introduce an approach of cluster-modeling for application on large road networks. We propose to first identify clusters of road segments that are similar in terms of attributes that are significant predictors of speed.