HDSTF: A Hybrid Deep Spatio-Temporal Framework for Traffic Prediction

Spatiotemporal sequence forecasting (STSF) plays a significant role in many fields such as climate, traffic, economy and agriculture. Traffic prediction is a typical example of such kind of problems which requires accurate traffic results in the near future to pre-allocate vehicles timely. However, there lies two major challenges in how to model the complex relationships hidden in spatiotemporal data. On one hand, the spatial dependencies are difficult to be captured and will change as time goes by. On the other hand, the temporal relationships are periodical and diversified, which is difficult to characterize. To address these two challenges, we proposed a novel Hybrid Deep Spatio-Temporal Framework(HDSTF) to solve the traffic prediction problems. Our experiments on two real-world spatiotemporal datasets demonstrate that our proposed framework outperforms the state-of-art methods.