pacific nw transportation consortium region 10 Pactrans

Enhancing Crash Classification through Attention-based Models: Unveiling Causal Factor Importance and Interactions for Improved Transportation Mobility and Safety

PI: Masoumeh Heidari Kapourchali (UAA), mhkapourchali@alaska.edu, ORCID: 0000-0002-2082-7879

Co PIs: Osama Abaza (UAA)

AMOUNT & MATCH: $40,000 federal from PacTrans; $40,000 federal Match

PERFORMANCE PERIOD: 8/16/2023 – 8/15/2025

STATUS: Active

CATEGORIES: AI, Safety, Transportation

UTC PROJECT DOCUMENTATION:

FINAL PROJECT REPORT: will be available once completed

PROJECT DATA: will be available once completed

DESCRIPTION:  The vehicle crash database faces an ongoing challenge of misclassified crashes, necessitating subsequent adjustments by researchers and traffic engineers based on reported circumstances. However, concerns surrounding security and privacy create significant barriers to releasing officers’ accounts, which contain invaluable insights into the conditions and circumstances leading to each crash.

The primary objective of this study is to enhance the classification of crashes by leveraging the potential of artificial intelligence (AI) techniques applied to event accounts and the statewide crash database. The ultimate goal is to develop, calibrate, and validate a robust classification model that can accurately categorize crashes, thereby significantly improving the identification of crash circumstances and causes.

By employing advanced AI and statistical analysis techniques, this research aims to delve into the language used in event accounts, enabling a deeper contextual understanding of crashes. This approach will facilitate precise classification by researchers or consultants, contributing to the identification of trends in crashes occurring on highway facilities. The outcomes of this study will provide valuable insights for enhancing transportation safety, promoting better mobility, and informing decision-making processes in the field.

Furthermore, the findings obtained from analyzing Alaska crash records will serve as a compelling case study, with the potential for application in other police databases within Region 10 and beyond. This broader application will support efforts t