Enhancing Traffic Prediction with Interpretable Community Embeddings via Louvain Algorithm

Predicting traffic is a complex problem that involves both space and time. This study focuses on the spatial aspect of this challenge, specifically how groups of road sections behave and interact within a city. Leveraging the well-regarded Louvain algorithm, we partition the urban road network into distinct communities. To augment the predictive power of models, we implement a learnable embedding layer that integrates generated groups with the input. We test our idea with a classic and simple model called Temporal Graph Convolutional Network (T-GCN). The obtained results highlight the promise of this avenue of research and emphasize its value for further investigation. Notably, the interpretability of the generated embeddings is demonstrated. By extracting meaningful relationships and disparities among communities, we provide insights into the dynamics of the road network. This approach enhances traffic prediction and contributes to a deeper understanding of the spatial interactions within urban road systems.