Evaluating the Robustness of Deep Learning Models for Mobility Prediction Through Causal Interventions

Abstract

Changes in the characteristics of mobility data can significantly influence the predictive performance of deep learning models. However, there is still a lack of understanding of the degree of their impacts and the robustness of deep learning models against the variability of these characteristics. This hinders the development of benchmark datasets for evaluating different mobility prediction models. In this study, we use a causal intervention approach to evaluate the robustness of deep learning models towards different interventions of mobility data characteristics, using both traffic forecast and individual next-location prediction as case studies.