Dynamic OD Matrix Estimation using Data-Driven Modelling under Data-Scarcity: an application of Gaussian Process

Abstract

Origin-Destination (OD) Matrix estimation in the transportation domain is a major challenge, often limited by data availability to estimate reality objectively. This state of "data scarcity" limits the reliability and capability of generalization of models, especially when simplifying assumptions are used to compensate for this lack of information. To address this issue, we applied advanced statistical methods that do not rely on prior assumptions, which are currently one of the most effective solutions to mitigate the effects of data scarcity. For this, Gaussian Process (GP) models have been developed and tested on a large dataset simulating realistic conditions to support this methodology. The test involved training multiple GPs with a training set that contained only 10% of the total dataset, which is the amount of data typically available in an operational and realistic scenario. The model was then used to predict the entire dataset, predicting key variables such as Departure Time, Arrival time, Activity Type, and Destination. The results demonstrate the effectiveness of the proposed method under limited data conditions and confirm the model’s ability to generalize to complex and realistic scenarios.