A Graph Signal Processing Approach For Real-Time Traffic Prediction In Transportation Networks

Accurate real-time traffic prediction has a key role in traffic management strategies and intelligent transportation systems. Building a prediction model for transportation networks is challenging because spatio-temporal dependencies of traffic data in different roads are complex and the graph constructed from road networks is very large. Thus it is computationally expensive to build and run a prediction algorithm for the whole network. In this paper, we propose a method to address these challenges. First, we introduce a novel spatio-temporal clustering algorithm in order to split the large graph into multiple connected disjoint subgraphs. Then within each subgraph, we propose to use a Graph Signal Processing (GSP) approach to decouple spatial dependencies and obtain independent time series in graph frequency domain. We make predictions along independent graph frequencies using adaptive ARMA models and later transform the predicted time series along each graph frequency to the subgraph vertex domain. Evaluation of our model on an extensive dataset of fine-grained highway travel times in the Dallas-Fort Worth area shows substantial improvement achieved by our proposed method compared to existing methods.