USING ENSEMBLE LEARNING WITH HYBRID GRAPH NEURAL NETWORKS AND TRANSFORMERS TO PREDICT TRAFFIC IN CITIES

Intelligent Transportation Systems (ITS) continue to meet major challenges in properly forecasting urban traffic under intricate spatio-temporal scenarios. This research introduces HybridST, a hybrid model including Graph Neural Networks (GNNs), multi-head temporal Transformers, and supervised ensemble learning methods such as XGBoost and Random Forest. This architecture concurrently captures spatial linkages, long-term temporal correlations, and contextual external information (weather, calendar, or control states). The model was investigated on multiple benchmark datasets—METR-LA, PEMS-BAY, Seattle Loop, LargeST, and V2X-Seq—and shown persistent dominance against conventional deep baselines like as LSTM, GCN, DCRNN, and PDFormer, with average performance increases of 8–12% in RMSE and 7–10% in MAE. Beyond performance, HybridST demonstrates high scalability and interpretability, making it appropriate for real-time applications such as: Urban mobility planning, Traffic management for large-scale events (e.g., the 2030 FIFA World Cup in Morocco), Energy optimization and CO₂ reduction strategies, Integration with smart city systems (IoT, V2X, digital twins). The study emphasizes repeatability, multimodal data fusion, and adaptability to developing-city settings in line with Morocco’s national mobility and sustainability objectives (NARSA).