Few-Sample Traffic Prediction With Graph Networks Using Locale as Relational Inductive Biases

Accurate short-term traffic prediction plays a pivotal role in various smart mobility operation and management systems. Currently, most of the state-of-the-art models are based on graph neural networks (Gnns), required training samples proportional to size network. In many cities, available amount data is substantially below minimum requirement due collection expense. It still an open question develop with small large-scale networks. We notice that states node for near future only depend its localized neighborhoods, which can be represented using relational inductive biases. view this, this paper develops network (Gn)-based deep learning model LocaleGn depicts dynamics aggregating updating functions, as well node-wise recurrent light-weighted designed few without over-fitting, hence it solve problem few-sample prediction. The proposed examined predicting both speed flow six datasets, experimental results demonstrate outperforms existing baseline models. also demonstrated learned knowledge from transferred across cities. research outcomes help systems, especially cities lacking historically archived data.