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

TL;DR

ALCo-FM is a novel adaptive foundation model that uses long-context multimodal reasoning and dynamic context selection to accurately predict traffic accidents, outperforming existing methods.

Contribution

Introduces ALCo-FM, a unified model with adaptive context windowing and multimodal fusion for improved accident risk prediction in urban environments.

Findings

Achieves 0.94 accuracy and 0.92 F1 score on urban accident data.

Outperforms 20+ state-of-the-art baselines.

Provides well-calibrated risk predictions with low ECE.

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 US 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 more than 20 state-of-the-art baselines in large-scale urban risk prediction. Code and dataset are available at: https://github.com/PinakiPrasad12/ALCo-FM