Enhancing Traffic Flow Prediction In The Presence Of Missing Data Through Spatio-Temporal Causal Graphs

Accurate traffic flow prediction poses a significant challenge in Intelligent Transport Systems. Most existing traffic flow prediction models operate under the assumption of complete or nearly complete datasets. However, real-world scenarios often involve missing data due to various human and natural factors. In this paper, we propose a novel approach, the Spatio-Temporal Causal Graph-based Graph Neural Network model (STCG), designed to address this challenge in traffic flow prediction. This model not only handles missing data but also automatically derives the causal graph, employing graph neural network techniques to capture nonlinear correlations between different sensors. It establishes a mapping between current and future traffic states, enabling predictions in the presence of missing data. Experimental findings demonstrate that compared to the benchmark model, the proposed STCG model yields superior performance in terms of mean square error, root mean square error, and mean absolute percentage error when data is missing. Additionally, the model significantly reduces computational complexity, thereby shortening training times. In conclusion, the STCG model exhibits potential applications in enhancing traffic flow prediction, particularly in handling missing data, thus improving prediction accuracy and efficiency. © The Author(’s).