MMCAformer: Macro-Micro Cross-Attention Transformer for Traffic Speed Prediction with Microscopic Connected Vehicle Driving Behavior
New transformer model cuts prediction errors by 9% using real-time braking and acceleration data from connected vehicles.
Researchers from the University of Central Florida developed MMCAformer, a transformer model that integrates macro traffic flow data with microscopic driving behavior from Connected Vehicles. It uses cross-attention to analyze features like hard braking frequency, reducing prediction errors (RMSE, MAE, MAPE) by 6.9-10.2% and shrinking uncertainty intervals by 10-24%. The model provides more accurate speed forecasts, especially under congested conditions, for proactive traffic management systems.
Why It Matters
Enables smarter city traffic systems that can predict and mitigate congestion before it happens, improving commute times and safety.