| Accurate traffic forecasting is crucial for the effective functioning of intelligent transportation systems (ITS). It helps in urban traffic planning, traffic management, and traffic control. Although deep neural networks have made significant progress in traffic forecasting, their effectiveness is compromised by noise and missing data due to sensor malfunctions, communication errors, and many more. While various missing data imputation techniques exist, they often apply pre-processing before prediction which creates an extra processing step. In addition, the spatio-temporal nature of traffic makes missing data handling more challenging since data can be lost at the temporal axis or on a spatial axis(missing data at multiple sensors). Thus, there is a need for a robust model that can inherently handle missing data and noise in traffic forecasting. In this paper, we proposed a Robust Probabilistic Spatio-temporal Graph Convolutional Network model that can handle noisy and missing data using the proposed probabilistic adjacency matrix and node-specific learning. Two real-world datasets with noisy and missing data are used to evaluate the performance of our proposed model. Our model surpasses baseline models in accurately forecasting traffic data, even in the presence of noise and missing data. The source code of our model is available at https://github.com/atkia/RPSTGCN. |
*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.