MMCAformer: Macro-Micro Cross-Attention Transformer for Traffic Speed Prediction with Microscopic Connected Vehicle Driving Behavior

2026-02-20 · Source: cs.LG updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Advanced, quick

Summary

The MMCAformer is a novel Macro-Micro Cross-Attention Transformer designed for traffic speed prediction, integrating microscopic Connected Vehicle (CV) driving behavior features with traditional macroscopic traffic flow data. This model addresses the limitations of existing studies that primarily rely on aggregated traffic data by incorporating individual human driving behaviors. MMCAformer utilizes self-attention to learn intrinsic dependencies within macro traffic flow and cross-attention to capture spatiotemporal interplays between macro traffic status and micro driving behavior. Optimized with a Student-t negative log-likelihood loss, it provides both point-wise speed predictions and uncertainty estimations. Experiments on four Florida freeways demonstrated that MMCAformer significantly improved prediction accuracy, reducing overall RMSE, MAE, and MAPE by 9.0%, 6.9%, and 10.2% respectively, compared to models using only macro features. It also reduced prediction uncertainty by 10.1-24.0% across the freeways, with hard braking and acceleration frequencies identified as key influential features, especially under congested conditions.

Key takeaway

For traffic engineers and urban planners developing intelligent transportation systems, integrating microscopic CV driving behavior data into predictive models is crucial. The MMCAformer demonstrates that this approach not only improves traffic speed prediction accuracy by up to 10.2% but also provides valuable uncertainty estimates, particularly beneficial for managing congested, low-speed conditions. You should consider leveraging CV data streams to enhance the reliability and precision of your traffic management strategies.

Key insights

Integrating microscopic driving behaviors with macroscopic traffic data significantly enhances traffic speed prediction accuracy and reduces uncertainty.

Principles

• Microscopic behaviors influence macroscopic traffic dynamics.
• Cross-attention can fuse multi-scale traffic features.
• Uncertainty estimation improves prediction robustness.

Method

MMCAformer uses self-attention for macro traffic and cross-attention for macro-micro interactions, optimized with a Student-t negative log-likelihood loss for point-wise prediction and uncertainty estimation.

In practice

• Incorporate CV data for enhanced traffic prediction.
• Focus on hard braking/acceleration frequencies.
• Apply MMCAformer in congested traffic scenarios.

Topics

• MMCAformer
• Traffic Speed Prediction
• Connected Vehicle Data
• Cross-Attention Transformer
• Driving Behavior Analysis

Best for: AI Scientist, AI Researcher, Machine Learning Engineer, Research Scientist

Related on AIssential

• Efficient and Interpretable Transformer for Counterfactual Fairness — FCorrTransformer and CAR enable interpretable, efficient, and counterfactually fair predictions for tabular data in regulated domains.
• Unveiling Stochasticity: Universal Multi-modal Probabilistic Modeling for Traffic Forecasting — Replacing a model's final layer with a GMM enables multi-modal probabilistic traffic forecasting, improving accuracy and robustness.
• DiffuJudge-AV: A Diffusion-Inspired Framework for Calibrated AV Video Evaluation — LLM/VLM judges for AV evaluation require calibrated uncertainty to prevent misleading safety decisions.
• Continuous Cross-Domain Traffic State Prediction via Memory-Augmented Graph Liquid Time-Constant Networks — MA-GLTC uses memory-augmented graph liquid time-constant networks for continuous, fine-grained cross-domain traffic prediction, improving accuracy in data-scarce regions.
• Geographically-aware Transformer-based Traffic Forecasting for Urban Motorway Digital Twins — Geographical awareness via mutual information significantly enhances Transformer-based traffic forecasting accuracy without added complexity.
• A Comparative Study of Graph Neural Network Layer Selection for Interaction Modelling in Driving Trajectory Prediction — ARMA, Chebyshev, and topology-aware GNN layers, combined with sum-based aggregation and multi-head attention, enhance driving trajectory prediction accuracy.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.LG updates on arXiv.org.