ALCo-FM: Adaptive Long-Context Foundation Model for Accident Prediction

Pinaki Prasad Guha Neogi, Ahmad Mohammadshirazi, Rajiv Ramnath

Keywords: Traffic Accident Prediction, Multi-Modal Framework, Machine Learning, Vision Transformers (ViTs), Graph Neural Networks (GNNs), Long-Context, Foundation Model

Abstract:

Traffic accidents are rare, yet high-impact events that require long-context multimodal reasoning for accurate risk forecasting. In this paper, we introduce ALCo-FM, a unified adaptive long-context foundation model that computes a volatility pre-score to dynamically select context windows for input data and encodes and fuses these multimodal data via shallow cross attention. Following a local GAT layer and a BigBird‐style sparse global transformer over H3 hexagonal grids, coupled with Monte Carlo dropout for confidence, the model yields superior, well‐calibrated predictions. Trained on data from 15 U.S. cities with a class-weighted loss to counter label imbalance, and fine-tuned with minimal data on held-out cities, ALCo-FM achieves 0.94 accuracy, 0.92 F1, and an ECE of 0.04—outperforming 20+ state-of-the-art baselines in large-scale urban risk prediction. Code and dataset are available at: https://github.com/PinakiPrasad12/ALCo-FM

Submission Number: 17