AI and Decision Support Systems for Crash Preventability PAR Processing
archive: archived pipeline: cataloged verified
Get this paper ↗ (full text — opens at the source; we link to it, we don't host it)
Summary
This report evaluates the technical, economic, and operational feasibility of using artificial intelligence (AI) and decision support systems (DSS) to automate the Federal Motor Carrier Safety Administration’s (FMCSA) Crash Preventability Determination Program (CPDP). The CPDP allows motor carriers to request the removal of crashes from their safety ratings if they provide evidence the crash was not preventable. Manual review of these Requests for Data Reviews (RDRs) and associated Police Accident Reports (PARs) places a significant burden on FMCSA analysts due to the diversity of state-specific report formats and the difficulty of processing non-searchable document formats. The study aims to demonstrate viable AI methodologies to reduce this workload and improve processing efficiency. The research team, led by the Virginia Tech Transportation Institute, developed a demonstration system using Texas PARs to test an automation pipeline. The workflow consists of eight steps: environment setup, PAR retrieval via web scraping, document parsing prerequisites, document parsing, narrative analysis, crash diagram analysis, eligibility/preventability analysis, and summary report generation. The system utilizes Python-based tools, including Selenium for data retrieval, Optical Character Recognition (OCR) via Google Cloud for text extraction, and Natural Language Processing (NLP) and Computer Vision (CV) for analyzing narrative text and crash diagrams. The model extracts specific data elements, compares them against state violation codes, and generates summary reports with eligibility recommendations for human analysts. The demonstration showed that the automated process could generate determinations for a single PAR in approximately thirty seconds. A cost-benefit analysis estimated that implementing customized parsers for the top 15 states (covering roughly 70% of submissions) would require significant initial development effort, ranging from 2,300 to 5,140 hours depending on the implementing entity. However, this investment is projected to yield a 75% efficacy rate in automated determinations. The study estimates that such implementation would reduce the total time spent by DataQs analysts by approximately 40 to 50 percent, saving between 854 and 1,020 hours annually. The remaining 25% of cases, often due to OCR limitations or insufficient data, would still require manual review but would benefit from the generated summary reports. The findings suggest that a "human-in-the-loop" model, where AI assists rather than fully replaces analysts, is the most viable approach. The report recommends expanding the automation from Texas to the top 15 states, building a repository of document excerpts for training, and developing a web application to improve usability. It also highlights the need for iterative improvements to address data loss and parsing inaccuracies. Ultimately, the study concludes that AI-driven DSS can significantly alleviate the operational burden on FMCSA enforcement teams, allowing for more efficient resource allocation and faster processing of crash preventability requests.
Key finding
Implementing an AI-driven decision support system for processing police accident reports is estimated to reduce analyst determination time by 40 to 50 percent while achieving a 75 percent efficacy rate in automated crash preventability determinations.
Methodology
modeling
Provenance
The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed. Discovered via bulk_ingest_rosap on 2026-05-23 (6 acquisition events logged).
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | rosap | — | — | 2 | 2026-05-23 |
| archive | success | — | — | — | 1 | 2026-05-23 |
| extract | success | cached | — | — | 2 | 2026-06-10 |
| clean | success | — | — | — | 1 | 2026-06-01 |
| chunk | success | — | — | — | 1 | 2026-06-01 |
| embed | success | — | — | — | 1 | 2026-06-02 |
| enrich | success | — | — | — | 1 | 2026-05-23 |
| promote | success | — | — | — | 1 | 2026-05-23 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 3 | 2026-06-10 |
| tag | success | vector_similarity | — | — | 19 | 2026-06-11 |
| verify | success | — | — | — | 2 | 2026-06-10 |
Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.
Topics
Ranked by relevance to this paper. Hover a topic for its definition.
Information type
What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).
- Empirical Findings: crash risk outcomes
- Methodological Resource: dataset resource