Graph convolutional networks for traffic forecasting with missing values

Abstract Traffic forecasting has attracted widespread attention recently. In reality, traffic data usually contains missing values due to sensor or communication errors. The Spatio-temporal feature in brings more challenges for processing such values, which the classic techniques (e.g., imputations) are limited: (1) temporal axis, can be randomly consecutively missing; (2) spatial happen on one single multiple sensors simultaneously. Recent models powered by Graph Neural Networks achieved satisfying performance tasks. However, few of them applicable a complex missing-value context. To this end, we propose GCN-M, Convolutional Network model with ability handle Particularly, jointly value and tasks, considering both local features global historical patterns an attention-based memory network. We as well dynamic graph learning module based learned local-global features. experimental results real-life datasets show reliability our proposed method.