Traffic Flow Prediction Based on Deep Spatio-Temporal Domain Adaptation

Accurate short-term traffic forecasting plays a key role in various intelligent mobility operations and management systems. Traffic flows have potential spatio-temporal correlations that cannot be identified by extracting the spatio-temporal patterns of traffic data separately. Furthermore, the problem of missing traffic data leads to the inability to train accurate models with sufficient data. Developing traffic prediction models with small training data is still a problem to be solved. In this paper, we study short-term traffic forecasting tasks and propose a method based on deep spatio-temporal domain adaptation. The experimental results show that our deep spatio-temporal domain adaptation model has better performance.